Data Science using Terno AI on FMCG Data

Key Takeaways

Hi everyone. Hi. Hi everyone. I'm allowing you to talk. Okay. How are you doing? Is the chat access allowed for everyone? I'll just check. analyst can. Okay. Hi everyone. While everyone is joining, while everyone is joining, feel free to introduce yourself. You're unmuted. Nish. Hello. Hey. Hello. How are you doing? >> I have muted you n I think. Hey Anand. Hey Bhavi. Hey Bhavani. Hey Jashwin. Hey Lokes. Hey Mitan. And Radhika. Okay, I am allowing you guys to unmute. Okay, I'll basically do that uh for next three more minutes. Okay, so be responsible when you are unmuting. Yeah. Hi Lkesh. >> Um hi sir, how are you? >> I'm good. >> This is Loc here. >> Hi. >> Yeah. Um I need to know about like uh what is the mo of this meeting like what we need to do now? M >> I I will basically as part of uh this meeting what I'll do is I would walk you through the use cases of turno in FMCG uh domain and yeah >> okay yeah like I'm working in an FMCG domain um the company name is double horse foot manufacturing company uh we are primarily focusing on rice based manufacturing >> okay Yeah. >> How big is your thumb? >> Um sir um per day uh we will be producing um more about uh nearly uh one to five tons in average sir um in in each products. >> Okay. Okay. Wonderful. Wonderful. Wonderful. Okay. So >> and this meeting is based upon like uh how how data is related to the company like is about data scientist like yeah >> correct correct how can how can data science help FMCG in general and how to achieve data science on your data on large scale that's the purpose of the meeting >> okay sir >> thank you >> wonderful wonderful to have your location. >> Yeah. >> All right. Anyone wants to introduce themselves, they can go ahead. All right. All right. So basically roughly the agenda I'll just share my screen so that everybody has the context. Yeah. While everyone is joining we can spend some time introducing ourselves. If you're not comfortable with uh unmuting mic you can put your your about myself in the chat. Would love to know you. I'll go through personally every message. So please feel free to put it in the chat. And yeah, so most of the people in this uh meeting are going to be from from from the FMCZ domain. Okay, those who have joined right now, I have turned on the unmute for them. You can Yeah, feel free to wonder, wonderful to have you here. Okay, so feel free to introduce yourself. uh feel free to introduce yourself using the chat or raise your hand and I can unmute you in case you want to ask a question. Are you guys able to hear me? [Music] Wonderful. All right. Hi. Great to have you here and wonderful wonderful to have you here. location is from FMCG sector and is working on rice manufacturing company and wonderful wonderful right great to have lots of uh yeah lots of great people in from the industry in this ing. All right, let everyone write to everyone. Okay, I've turned on the chat. Okay, Prince, can you hear me? Uh, now shall we start? Wonderful. I would love to have your intro introduction everyone. Okay. The the core idea of the meeting is to go over the use cases and the case studies of data science in FMCG domain and also talk about how you can achieve that using the uh tero that is the main main idea and objective. So wonderful, wonderful to have everyone in this meeting. Great. >> Hi sir. >> Yeah. Hi Lkesh. Go ahead. >> Can you explain like what are the use cases that you need so that we can able to explain even more like >> Yeah. Yeah. what are the >> I will go through the use cases that we have right now and >> okay >> then then you'll get a picture >> okay thanks for that yes >> go ahead yes you've raised the hand please go ahead can unmute you have the I have uh allowed you to unmute Okay. All right. Feel free to raise your hand. I'll turn on the unmute for you. Okay. Okay. Sure. Sure. I'll I'll What happened to my profile? Okay, that's about myself. I am Sep Giri. I am the founder of Turno AI. I have previously founded cloudex lab and tits global. I am from the engineering background and the data science. I have worked on the large scale computing at Amazon in mod shop and devre recently and I graduated from it. So but I have been in the industry uh for quite some time. I have um I mean something that I don't mention here is that I founded a a home delivery food system in Hyderabad that I ran for about 7 months and that's uh my um touch with the the real consumer market and uh but most of my life I have been into engineering and data science across different companies. I and uh I have worked with I've built the companies like Freebits where most of the customers were uh large companies most of the power plants in India are built using my software. All right. So now right so the agenda for today's meeting is uh basically the okay close to all right that's me you can you can search on LinkedIn and you can find me I'm Shep Giri okay all right so this is the agenda so we'll go over the FMCG landscape in in like briefly Then we'll talk about what exactly is Turno AI. Then we talk about the use cases and then we talk about the case studies. Now with respect to the case studies uh we will go over these four case studies out of all the case studies that we have built around turno and you can take a look right just before I get started. So Turno is being used at one of the top FMCG and and pharma companies based out of India but they are a global multinational. So so I mean the most common products the one company that manufactures one of those most common products in India is using turno for their marketing and sales data science. Okay. All right. All right. Wonderful. So, no questions. If you have any question, please put in the chat or raise your hand. I will unmute and answer your question. All right. So basically the FMCG landscape is it's a fastm moving consumer goods industry and these are the every organization under FMCG is experiencing an unprecedented transformation. On one hand there is a behavior change in the consumer and on the other hand there's a digital disruption. On one hand the consumer have become very price sensitive. On the other hand consumer has got more uh disposable income. So they they are ready to buy in new products but at the same time they have become very price aware. they can compare everywhere. Similarly, there's a digital disruption that's happening and and and also the customer behav consumer behavior with respect to buying is changing drastically. For example, the the the consumption of the um the cons consumption of various uh unhealthy products is going down in recent past and and so on. So there's a lots of lots of change in the consumer behavior because of the media, because of YouTube channel, because of uh the changing economy and and therefore it is important for every company to keep the pace with the consumer the the evolving behavior and essentially it's probably it's a it's a pretty big market. It's a 15.6 uh trillion dollar market and it's going to reach about 18.5 by 2028. And the companies that are that are really doing well are the ones that are datadriven. They're trying to figure out what variant of the product sells good and what doesn't. Okay. And the AI in FMCZ is going to be a key differentiator between successful company and nonsuccessful company specifically the innovations related to the data science. Okay. So because in case of consumer goods you have to understand the consumer behavior on a large scale and you have to analyze the history of your products and so on. And so and then inventory costs saving is another very very powerful utilization of of the um data science in in FMCG. Before I go about turno, I want to just walk you through the case studies that'll help you, right? And those who are in hurry, they can basically just go to turno and sign up. You can go to turno and click on sign up and you can start using and experimenting with it. This will give you a quick experience of DO. Okay, I'll just skip this part. I'll come back to this again. Right? So I want to take couple of examples right now because all of you must be in you know um must be having all of you must be having a question that how does this event relate to us. Okay. So if you are from FMZGZ domain you will basically be able to relate to these uh examples. So here's a hypothetical company called wellness co. Okay. So turno segments global consumers into three groups. Say say for example this wellness co is launching a new natural sweetener. Okay. And they need to launch launch this new natural sweetener and figure out the marketing plans for it marketing and the sales plan for it. So what they do is they ask uh turno they give all their data about their consumers. The wellness co gives and connects all the data sources to turno interno and then they ask uh turno to segment the global consumers and the interno breaks down the the consumer data into three groups. one say health conscious millennials household with diabetic members and and trend followers and figure out figure out which users are where in what geography. Okay. And once that is done then you can basically create the personalized marketing by region by group of uh by region by city by country and by economic growth and by channel. Say for example you realize that your target customers are mostly on Instagram and Tik Tok in these areas and so on. So so essentially essentially once you can segment and understand your consumer base globally you can come up with the marketing plan. Now for the companies that do not have their own data, they might want to buy the data from the data vendors say for example Neielson and there are quite a few companies which provide you a lot of data about consumer behavior. So if you are launching a product and you don't have the data you can buy the data from lots of vendors and then connect the data to turno and you can ask turno to do all kinds of analysis. All right. So that's one use case and you can clearly see that okay um in this in this geography we should try retail loyalty in right in in this geography you should try Instagram and Tik Tok and so on. So your your comp campaigns are going to be very targeted. This is just an example. This is just an example. I mean the e-commerce companies have similar example whereby they can also include say email. they can include you know targeted ads on their own platform like in case of Amazon and so on. Now here's another use case. Say for example, Pure Life products a global FMCG brand, right? And they want to they they are produc they they produce uh P pure life produces protein drink mix. Now this drink is seeing seasonal surges in Europe during January as customer adapts fitness resolutions and Turno detects this pattern early. Okay. So based on the forecast, pure life adjust its distribution network. More shipping containers and trucks are rooted towards European market and warehouse in Southeast Asia are scaled down temporarily. The outcome is 8% reduction in storage costs and better availability of products in high demand regions during peak season. Now this is a real you can say operational challenges in FMCG, right? the demand and demand basically ba basically even in the quick commerce case uh in India one particular store might be seeing much more demand while other might be seeing uh less demand at one particular time. So you can do this demand prediction you can do the demand prediction and and or or maybe anomaly detection in your data from the past history and give the results. So basically you can decide upon your operations using the using the insights from terminal [Music] and moving on right next is the supply chain and operations okay so let's take another example called global health inc a multinational FMCG producer right Factories in Brazil, Germany and USA use IoT sensors to track ingredients usage in real time. Now turn highlights that Brazilian plant consumes 15% more cocoa uh cocoa per batch than the global average. Okay. So basically based on the data you can come up with these insights that optimize your operations. Okay. And after the fix so the investigation reveals the calibration issue with one machine and after the fix annual cost saving reach to about $2 million. Production becomes more efficient with less material waste and so on. So you can even predict the the operational challenges. Let's say one particular machine is not performing great. If you can look at the data and look at let's say the deviation and so on. So that turno will do it by itself. You can ask turno to do the analysis do the do the do the what do you say anomaly detection or do other things. It will do that for you and give you the results that are needed. Okay. Other use cases could be um basically yeah this is a diagram representing the same. Now imagine that there is a company called Bioare International launching innovative wellness product. Global trends analysis by turno identifies a rising interest in gut health and probiotic and essentially bio bioare prototypes um new yogurt called probiotic yogurt and tested globally and customers testing when we when we tested it on the customers it shows that there is a 80% acceptance rate in Japan and South Korea and only 35 acceptance rate in North America. Right? So now basically using the data using the data we can figure out which markets are great for a product. So yes, so here is u basically a example that talks about AB testing and and optimizing by doing the analysis of AB test data. Okay, I hope it makes sense. Okay, the use case number five is about let's say there is a company called Green Life. It's an FMCG company focused on the health product and Green Life integrates with fitness apps such as Fitbit and My Fitness Pal. Okay, so Turo analyzes the users data and send personalized s product suggestions. Consumers with high sugar intake gets nudges like try our zero sugar chocolate bar. runners are recommended protein recovery shakes for uh faster recovery. This shifts Green Life's image from product manufacturer to a proactive wellness partner boosting long-term consumer engagement. Okay. So this is just an example that bas based on the users's data we we users data and the steps basically gives the suggestions like zero sugar bar or protein shake. Okay. Now there's another company Pure Essence. These are hypothetical companies because we wanted to remove our customers name in in this. So, so Pure Essence, a global skincare FMCG, right? Right. So, Turno builds a price elasticity model and finds that if you cut down half a dollar on the skincare gel, right, the sales increased by 20%. in Latin America in western Europe demand is largely inelastic. So let's say you have a brand that you have been selling for some time right a common sense will say that in decrease the price and observe and you know make money but I mean make more sales and hence right but it's not necessary that dropping the price will give you more money dropping the price will drop down your profits and sometimes dropping the price can also drop your demand So essentially to do this analysis whether dropping the price helps you make more money you need to look at the data you need to look at past when you gave the discounts did it increase or did it not increase did it increase for all the users or did it increase only for a certain geography okay did it incre dropping the price impacted what all products or what all variants of your product, right? Based on these insights, say pure pure essence adopts to a region specific pricing strategy. Okay? So for for example, comparative pricing in emerging markets while premium pricing in the developed markets or developed countries. So this is how I mean if you have to plot it and so on. Now coming back to it. Next, let me take the another example of quality and compliance. Take the example of a health plus cop. It's an international FMCG company. Health health plus deploys AI powered vision system on packaging lines in the US and Europe. Turno automatically detects defects such as crackport bottles or mislabeling in real time. The benefit includes 25% fewer consumer complaints, improved production speed, and consistent product quality. This reinforces brand trust and reduces manual inspection costs. Okay, the next uh use case is in right it's the same use case essentially. Now before I go on to the case studies where I'll walk you through uh let's say thisa case study whereby in this case study we figure out if discounts are killing the profit or discounts are good. So I will talk about this case study in much more details and this is how this is how the case study looks like. This is a this is the result of interaction with turno. Give me a summary of data set. Give me the visualization. Turno gives you this data. Similarly, visualize sales versus profit by category and subcategory. You get this result and so on so forth. So you can see here. All right. Now, now right before jumping into case studies, let me step back to what is Turno. Okay, so Turno is your AI data scientist. Okay, feel free to ask questions before I Yes, Bony. Let me open. Yes, we you will get a a copy of this presentation. I will be we'll be sharing with you. Okay. All right. Shall I unmute you? Uh Bavan. Okay. All right. No questions at this point. Hello. >> Yeah. Go ahead. >> Good evening sir. >> Yeah. Go ahead. Hi >> sir. Myself Ban. I'm studying engineering student. >> Okay. >> So I'm aspiring data scientist. I thought this session would be helpful more helpful for me to become a data scientist and uh this turn on and how to use >> to building the um insights and also more attractive visualizations data visualization. >> Yeah, >> thank you for on sir. >> Okay. Okay. So yes, I mean to become a data scientist, you need to understand industry, you need to understand use cases and how it impacts and I think this session would help you in that sense. Thank you. All right. Any other questions before we go ahead? Okay, I've clic clicked on your name immaturely. Okay, so yes, uh go ahead if you any anyone has any questions, I would love to answer those. Okay, a question from Loesh is that uh project can you provide is there any use case uh project can you provide? Yes, do reach out to us. uh right some of the data scientists who are working as intern with us we might I might connect you with them so that you you can basically work with use and get insights with you okay we'll do that all right drop me an email at reacheso that'll be great okay a question from custo is is an end to end workflow for a real world FMCG company a part of presentation so uh basically the the way it works is there's no end toend workflow okay there's no end toend workflow each pro each project has got a real end toend workflow so I'll walk you through four examples of end to end workflow of uh the various case studies from the FMCG company. Okay, that was Cost's question. Any other question before I move ahead? Yes. No questions. Moving ahead. So, so far I talked about some of the use cases of AI in FMCG. Okay. The simple essence was if you have data, you basically try to get insights from that data. It or you can combine your data along with the data vendor say Neielson or somebody and then get the results and do various kinds of analysis in terms of operational excellence, marketing plan and sales strategies as as well as anomaly detection and other things. So that can be quite quite useful in industry. So what exactly is turno? Turlo turno is uh basically your data scientist. You ask questions in plain English. It responds in desired format. Okay. So you basically ask questions, it interacts with your data and gives you a result. Right? The data can be anywhere. Anywhere practically anywhere. Data can be in your databases, ERPs, uh, in your data links. Data could be in your files, PDF, Excel sheet, CSVs and so on. So you can or maybe it could be part of let's say third party APIs or it could be part of the internet. So you can basically connect any kind of data source to turn. Okay. Now the core engine of turno that connects to the database there is something really unique about turno the when it's talking to your database it has enterpriseg grid security it is this is the most uh important part of kernel the SQL shield so your data stays in your organization your data never leaves your premises that's the most important part. And you can have the you can hide some of the some of the you know you can basically you can give the rolebased access control in your organization. Some some of the team members have access to one one one table other members have access to another table per column per row. you have all sorts of security in place. Okay. So the basically you can do exploratory data analysis, advanced AI models, realtime processing and automate automated report generation using terminal. And how does it work is user uploads a question, user uploads a file optionally if you have extra data from somewhere in terms of a file, video file, PDF file, Excel file or or maybe image, you attach that to your question, go to turno, interact with u basically turno interacts with your data, writes the code, writes the SQL, your data could be inside something called model context protocols, AI uh services, databases, files, you know, your CRM, your ad networks, third party datab databases, everything and then turno gives the answer back to the user and turno takes multi-step process. You give a question, it comes up with ideas for each idea. It tries to solve each idea, comes up with more idea and it keeps on coming up and solving until the answer is given to the user. Yes, Amit. >> Yeah, go ahead. >> Uh, hi Sep hope you're able to hear me. >> I can. >> So, so yeah. So Si just wanted wanted to understand so this in this diagram that which you just showcased that uh that user that needs to upload files plus users question to your turno AI right so can I understand this turno AI is kind of an engine that what which basically talks to all those datas and other things like how because you mentioned like interacts with data using code or SQL is it some kind of engine that which you people have prepared like >> so you can you can see my screen >> uh yes So the way it works is that you basically just sign up. Okay. And >> yeah, my machine is overloaded. Okay. So you can basically sign up and you can start interacting with it like you go like here, right? You can basically uh you can just ask question. Optionally you can attach your files. Okay. >> Okay. >> Mhm. >> Optionally, by default, I have configured three databases here. One is a demo SQLite database which is >> this is a standard demo database. This is global data like global literacy rate, global internet usage, global inequality. >> This is the datab data from ERP. This is the ERP database. >> Mhm. >> And it's loading just too many tables. Okay. So yes this is uh essentially you know the in >> so these are all like kind kind of a dummy datas or there are or these are kind kind of real datas um >> we can't share the we can't share the real data so we have put in the >> dummy so that's what I'm asking so this seems to be a dummy data right or not from a not from a real data perspective >> essentially this is there is a product called odo just like SAP there is a called ODU they make they have the very popular ERP. So >> okay basically they have created a dummy data source where they okay dummy >> so this data okay fine so this data belongs to the Odo organization that who has who has published you to utilize in your terminal right if if I understand correctly right >> exactly >> okay okay >> so basically your you like you your organization might be having more hundreds of data sources okay you basically connect all your data sources from >> Oracle, MySQL, Postgress, BigQuery and so on, right? And in case you have four data sources, we can connect them by the way of something called MCP. We we have a philosophy called MCP with which you can interact with anything. You have Shopify account, you have Google Drive, you have Microsoft, everything can be connected at some place. So you so turn knows about your data. It is studies and tries to figure out what is the meaning of each data set and makes notes of it >> and then when you ask questions it basically does the analysis in detail. Okay. So when you ask uh let's say here here >> Mhm. here analyze leads and and want deals by source campaign medium using UTM campaign and this so I just told it a little bit more about my data set >> in your organization once we set it up we don't need to explain any of this okay any of this it'll basically understand automatically and then we'll continue to use that and then here >> it basically starts the analysis and and Then it comes up with these results in the end. Okay. >> Okay. >> Yeah. So it it shows that from the search engine we are getting these many leads. These many of them were converted. This was the revenue. This was revenue per lead. This was the lead recall and so on so forth. You see that? So, so in this case any any size of data or any kind of kind of a quantum of data that needs to be fed into to turno to analyze or like uh how exactly? So even pabytes of data it's able to do there's no limit because we are not uploading >> we are not uploading your data to JPT or anything right >> okay >> the way it does is just like a data scientist would analyze the data it follows that strategy you ask a question what will data scientist do it will just go and probe your tables it will understand your problem statement it will look at the tables understand your data structure and then we'll write code whether in Python build models build all of the things needed in order to answer your question. So it doesn't just take your data and upload somewhere. No, your data remains in your organization. So what >> how do people use Turno is how do most of the companies use Turno is they install Turno inside their cloud. Okay, this is this is only for demo and for small clients. But for real big clients, they they they can't they can't share the data with us. So they take >> install it on their cloud inside their own uh servers and their data is already there and turno talks to their data, studies it, writes the code, tries this, tries that and gives you the answer eventually. Okay. >> All right. All that means we need to install this uh turno AI in our system and and this uh turno will internally talk to to all the data sources and then uh publish all those results. Uh is that correct? >> That's that's correct. But those who do not want to install it right now they will they basically can just come to app I mean turno.ai I sign up and start using. >> Okay. And that way they don't have to set it up and so on. But for you when you are connecting your own data, you will have to allow turno servers to talk to your database because your database may not allow us to connect if you're using our servers. >> Yeah. >> Lost you. >> Uh let me repeat. >> Ah yeah, please. I think I think I I think we lost you I guess for a few seconds. >> Okay. So let me repeat. So most of the large customers who who do not want to share the data outside their boundary what they do is they take turno and install inside their >> correct organization. >> That is great for many of the big companies whose data who don't want to you know share the data with us. Some of the companies which are small a smalls size organization what they do is they basically connect their data sources here after signing up at turno.ai and then when you connect here you'll have to go to your database admin and allow our servers to access your data. So it'll give you the IP addresses so that we can access otherwise we won't be able to connect to the data. >> Correct. Correct. Correct. Correct. Correct. Okay. >> Very wonderful questions. Wonderful questions. >> All right. >> Sundep, quick question here. >> Yeah. Go ahead. >> Yeah. So the question is like uh if for example I install your turno AI on my cloud. >> So now it will be using all my uh server uh servers only. >> Yeah. >> Uh as per the requirement. >> Yes. And if I connect all my data sources. >> So as I understand will it solve uh the uh the two things like one is uh if I ask the uh ask uh the questions are around uh different analysis analytics that is kind of a business questions if business we give an interface to business. >> Yeah. >> Interacting with turno AI >> can we also integrate it uh in a similar way a kind of a chatboard also like chatboard interface. we have some chatbot interface and at back end uh can we integrate turnover as well for the uh for answering the question of the customers. >> Uh good question good question. So we are building we building the Slack bot we are building the Microsoft teams uh bot as well as WhatsApp bot. So we building those but we are open to building other bots and the only thing is it is designed to be data scientist. It is designed data scientist right and it ex like because customer will need the instant answer. Data scientist takes time does detailed analysis and then gives the result. Right? You may not want that to happen with the customer. Right? So in those cases what people usually do is they after they finish the whole analysis what they do is they create something called something called apps. Okay. In some cases it would give an suggestion that that you can create an app out of it. Okay. So, so you create an app and once you have the app you can basically launch it uh you can just use that app to answer user question. Okay. So customer we haven't thought of that that like can you use it for customer? >> Yeah. Uh see my thought was because you already have this entire um uh ecosystem where you can connect uh it with with different data sources because data source for business will be different and data source for customer will be different a little bit different there will be transactional uh database those are common for both. >> Yeah. Uh uh so in that case um if it like it um if your ELGO is working in a way where you do all the analysis and put it in a reporting way for customer which I can see in the demo also. >> Yeah. >> Do you have a readily available u uh elgo for answering the simple questions? Um like see it it just came in my head like you might be having two set of algos and one is working this way the other one is working with answering the questions in a simple way not doing uh like this kind of analysis. So is that uh also available with you like something like that? uh we basically what we have done is uh right you can here you can customize the prompt the behavior by the way of talking to it. So you can ask it not to do a deep analysis instead just do singlestep analysis and it'll do only the singlestep analysis that is there but again I need to think about whether you can expose it to your customer because it is designed like a data scientist and usually data scientists are not that polished when uh it comes to interacting with your customers. Can your algo behave differently si on on different variety of other datas like uh like x data or y data. So do you think that your algos will behave differently? >> Yes. >> Or it will it will take uh everyone into same consideration. >> Yeah. Yeah. Okay. Like for example here I said here is a resume of someone. Okay. >> Okay. >> Extect the text from the resume if this is suitable for DBA or not. Okay. So I just want to check if this person is suitable for DP or not. So here it is studies and it studies the resumeum and tells that the rumé shows some datab based exposure. Okay. And right conclusion is the candidate has basic SQL and ETL familiarity. They do not demonstrate. Right. So this is completely a very different task. This is not what we designed the tero for but it does the job. Okay. So, >> so the statements are like hypothetical statement or they kind kind of or kind some kind of some kind of imagin. >> This is a real resume. >> Okay. This is a real resume and for for even uh ourself if we are reviewing someone's resume I I use this >> okay it's real everything is just real if you you know it has done a very good analysis >> of the resume >> okay >> so uh yeah >> so this happens only one uh sorry to pause you know sep which you are talking about so can I feed like number of like n number of uh profiles together and it scans and provides the conclusion one after another or like it will like like only one at a time like >> no it can do everything I'll just show you something okay >> uh right okay so I may have to quickly close it because it's like uh confidential data okay and I just uh right say for example uh find customers here I'm using using this thing to do sentiment analysis of things okay here I said that right could these sentences horn a graph such that things which are similar they come close by okay so it has drawn them >> I want to eat pani puris. Does freely exist? Does free will exist? Can a machine become conscious? So does uh like here we love India. Bharat is a great country and so on. So it has put them close by and Android and iPhones are useful. So it has put them close by and so on. So so it can do all sorts of analysis in much more details. But I I wanted to show you something. Um there was an invoice example where I took the data from all the invoices and put them together in an Excel sheet. Okay. Okay. Yeah. Like here here I'm trying to do something very different. Okay, split the attached video into two equal length videos. Right. So, it has done that splitting part. Finally, it has done the splitting part. Okay. And it has put it in my uh files here. This is this the working environment for this data scientist. It keeps on preparing all the things uh at one place. Like in one case I converted a lot of PD apps into an Excel sheet by combining and extracting all the valuable information. It is there somewhere. I'm not find that. Yes. So this is the exercise I gave it here like these PDFs containing is each PDF contains the table of work done. combine all rows from the table into CSV. So this is the these are the files they gave no analysis and then >> in a single shot >> in a single shot. Yeah. >> Okay. >> And after get finishing the work it created a nice CSV. It studied multiple times. It tried uh one strategy that did not work. Then it tried another strategy that did not work. Then it tried another strategy that worked finally. Okay, because the task was complex. So it basically does all of that. Okay, you can see that it is trying to do all the installation and everything and yeah and finally it gave me uh the result somewhere. Oh okay. So it basically saved it here. It did the job. Yeah, this one here it it has done the job for me. Okay, this it has combined the result from all the PDFs into one. Okay. So you can see yeah you can see it has combined the jobs from everywhere. Okay. So that that's something you know uh it is versatile. It is not necessary that it comes up with certain set of steps based on the problem what problem it is trying to solve. It would come up with the ideas and solve it. Okay. Similarly, imagine that you have some data in your one database, some data in another database, some data you have to download from internet and so on. It'll do everything and give you the results. Like for example, one of the clients in FMCG, they want to do the forecasting. They want to do the forecasting of uh of their product. So what it has done is it gathered for every city for every city in the in India it gathered the climate monthly climate as in the temperature the humidity and and so on there three four uh parameters it has gathered and then it has al also gathered the education level from the internet for each city and and basically then it took took the data of every city like how much do we sell in each month historically for 10 years and then over overlapped the two data and then did the prediction then it did the prediction for the weather for next month in that particular city where you want the prediction and then it predicted how much demand will be there. So it comes up with all these ideas that are something very uh you know something which um take which takes like months and months of time. So you can overlap all the data and you can ask it's it's just a matter of you know how you ask questions. That's all. Hey Sep go ahead sep one one question sep the data that which you're just talking about so can we have some kind of a data which is on the based on the kind kind of some kind of a real streaming data on a live basis data can we do that can we have that >> or it is just some kind of data that which is I'm sorry >> connect that using the this part there is a philosophy >> kind of a live streaming data that probably we can we can make use of and uh probably then we can uh like um >> yeah sorry >> one one question I do have a question uh Sundep like uh uh like see every business is unique >> yes >> and every business has uh a unique data set and like it's the the verbias the words we use in any business so it's different you know So how will you like uh if like my data my data sources are integrated in turno? >> Yeah. >> How turno like is it uh like your algorithm will take care of everything? Uh we need not required to like uh train it on uh our uh uh like the the our business specific words uh or business specific definitions mean how how turn will understand that like if I'll be asking question any specific that is business business specific questions so what about that >> very good question and that's a question coming from uh real uh experience. Okay. Um so I think I couldn't I couldn't relate with this question um two months back or three months back but I can relate with it and for that okay so for everyone else the question is every organization has their own lingo own language own slangs right and how do you deal with that with uh you know turno right you do you What do you do in that case? Right? So what we have is something uh one is basically there is something called organizational prompt. Okay. So you can essentially just create a organization prompt whereby what you can say is in my company people call um you know this product as this this product as this and so on. Right? So for if let's say a company is making sugar-free something like or Coca-Cola right we call it like CC for Coca-Cola and things like that right so you can define all of that inside organization prompt that is the first level of customization right now all of this the way we do is that when we are implementing for our organization or anyone is implementing for the organization what they do is they Hey, here is the question that person asked. Here is the uh SQL query or Python code that they wrote to answer this question. Learn from it. So, it basically learns from your past reports and comes up with all of this uh hidden knowledge in your organization. Okay. So, one is you manually write the prompt. Secondly, you let the turno figure out the the local language by giving it the historical data, historical reports. Uh and then it basically comes up with all these ideas. Third way is that when you talk to it, you say that in my organization as SF is called sugar-free. So it basically learns from it it it learns from that. Okay. and and next time when you say SF it will always understand it to be sugar-free and things like that. So there are three ways to do it. One is a manual prompt curation where you define your organization specific knowledge. Second is you give it past reports and past solutions. It learns from that and third is when you interact with it, it'll keep on learning with you. That's the >> so you mean yes mean that second part is we'll have to upload all our previous different kind of reports that we have generated or used something like that. >> Yeah. >> Okay. Okay. And uh uh like if because if you share your uh business lingo, you share your business definitions sometimes >> uh it it it uh uh there are some um like business specific things which we do not want to share and it automatically goes with data. So in that case >> it will not come here. It will be in your premises data right will be installed or no it'll be in your organization and all of that data will remain with you >> right right understood but okay so you're giving this uh so it is secure if I uh installed in my landscape in my on my servers >> uh turno do not have any data collecting things in the background or anything or sourcing it somewhere else it's not like that you know >> that's right that's right >> okay it's any any platform that we purchase and use install in our environment it's similar to that >> yeah yeah on top of it what like the one very important thing it does is in case you want to mask certain even the meta data you can mask it by renaming it okay and you can also remove multiple columns from the tables right you can delete without touching your actual database. Let's say there is a particular table that contains people's credit card information. You don't want analytics to touch that, right? And you can't change uh you can't go and ask uh DBAs to change your existing servers. Remove this column, remove that. In those all cases what we do is you know you you go to the admin go to the column configuration remove the mark many columns as private and turno or llm will never touch those columns. Okay. So uh we provide an extra layer of security on top of your data other than database already has security on top of that if you want to further put the put the wall uh you can put it using terminal. And what do you think like what is the estimated time that it will take in getting trained up to a mark where it starts uh giving a proper um uh like the reports uh like uh yes >> okay so so usually it doesn't take much time right it'll basically it basically uh is pretty efficient at getting it done the only challenge challenge is when you have tribal knowledge when you have local knowledge which is not noted down anywhere like for example there's a company who uh bought recently they are using turno now they had acquired multiple products uh in past so every product data is in different database and from after certain date the data is in one table after certain date the data is in another table so This knowledge that for this from from this period to this period go here this period only this kind of knowledge is needed to be codified and but by default as soon as you connect 90% of your questions would be it would be able to answer just just by connecting without providing it any context. Yeah. So it should be I mean the more you work with it the accuracy keeps improving on day one you'll get 90% of your results >> isn't I don't know I have an echo >> okay yeah can you try Huh? Everybody's on mute. So can try now. I'll try to be on mute. Oh, go ahead. Okay. All right. Uh, any other question? Thank you, Rohan, for answering those questions. Okay. Yes. Go ahead. Liz, can you try unmuting? I'm turning on mute. Go ahead. Yes. Go ahead everyone. No questions. All right. >> Hello. >> Yes. >> Am I audible? >> Yes. >> Hello. >> Yeah. >> So, where will you share this PBD? >> We will share over your email. Okay. >> Email. Okay. after the after the after this right? >> Yeah. Yeah. You can leave your leave your email in the chat as well. >> Okay. Okay. >> Am I audible now? Nan. >> Hi Nan. Yeah, you are on. >> Yeah. I have a small question. As I understand is this similar to like a chat GPD kind of model but say uh it kind of customized to our company requirement where it can access companies internal data and ask any question with respect to data right >> that sure what it is doing >> yeah one big difference n you can see my screen it's a good >> right so question is how is it different from chat GPT right >> correct >> so main difference is that as you that you understood it correctly that it is inside your company it doesn't upload the data to chat GPT or anybody instead >> correct >> it comes up with ideas it write code it writes code it writes SQL it does all of that okay inside your organization and generates the result it only shares the information the names of the columns names of the metadata that's all it doesn't share your actual data with chat GPT or anybody >> one followup question for example is it if there's a standard method that the company has already defined for doing a customer kind of segmental analysis >> yes >> and if I feed that methodology in this uh termo tool >> then termo tool will be able to do that standard analysis at a defined frequency that I would have feed it right and give the output on that day >> of course >> all can be done. >> Yeah, even if you have not defined then also you'll be able to do. If you have defined it's well and good. You'll be able to do that in your organization. >> Okay, got it. Great. Thank you. >> I think I think I resolved my microphone issue. So here goes my question again. >> Um so so it learns, right? It learns from from data >> that you already have in the company, right? So I'm working on something similar but conceptually not building a tool like you have right where every client basically let's say the example of the customers where you were talking different cities different education level it's like every client gets vectorized >> can you remember that threedimensional thing you showed us where like similar topics get bundled together so similar clients get bundled together and that's how you predict okay because you know these guys behave similar those behaving. So it's like a segmentation process that happens within the tool. Is that what it does? Is that what Termo does? Like bundle, it can bundle similar. >> No, it's not. That's not how it works. >> Okay. Okay. >> So for every problem it comes up with hypothesis you want to do let's say customer segmentation, right? It comes up with a hypothesis that where is my customer's data? I will first get the customer's data then I will put it in this format then I would apply this algorithm right it comes up with these hypothesis and then for each of the hypothesis it comes up with algorithm it comes up with the code Python code and then it runs it checks the outcome then again does next thing and next thing and so on so it pro it is exactly the way a data scientist would do if you ask me to do something. I will come up with lots of hypothesis. Against every hypothesis, I would figure out which data sources I need to get the data, what code need to write. I run it, then check. I run it, then check, and then once I get the good result, then I give it to you. It works like that. >> Okay. Okay. >> Right. So those if those who are curious, you can take a look at something called chain of thought and actions. Okay, that's what Turno does. On top of that, it also has a very big component called uh learning with your system. We call that semantic lay uh knowledge layer. Then it has a semantic layer whereby it has its own new kind of SQL engine. What it does is you write the SQL and the questions that you're asking or product that you're asking could be um vague right you can say select start from products where the product is um something that I wear on the top right instead of saying shirt I'll say something that I wear on the top so it'll basically just match that so it has all of that engine we call that semantic engine which basically gives the semantic capability to database. On top of it, it also has a secure engine which secures your database and gives the result and that's what it does. Okay. >> Okay. >> So, uh yeah. All right. I want to walk you through the use cases of turn. Go ahead. Nan, raise the hand. >> Yeah, I already asked my question. Thank you. >> Okay. Wonderful, wonderful, wonderful set of questions. I'm pretty happy to see amazing questions. >> Hello. >> Yes, dep uh by any chance is this recording going to be available? I basically missed it >> because I have >> You'll be sharing We'll be sharing the recording. >> Yeah, sure. Uh another question I was uh curious about. I'm new to uh all this AI stuff that I'm trying to explore. Uh you said uh sometimes back that u uh about the chain of thought and you know the engine identifies u it basically tries on uh gets on a hypothesis uh tries writes code for each of the hypothesis steps and then tries to identify whether each step has satisfying output or not. How does it actually understand what is what is a satisfying output? >> Yeah. So basically it I mean most of the LLMs understand simple question that here is the result here's the question it can whether it answers or not. So you have like a multi- aent system, one agent working, other agent verifying. So all of that is happening in parallel. >> Yeah. But then verification is entirely uh u I mean it changes, right? I mean let's say what I'm saying is that what is the weather today >> in India? >> Yes. >> The agent says the weather is X but then it might be correct. Then still the other agent who is observing says it is not correct. Yeah. So, so in 90% cases, right, both cannot be wrong. Okay. So, if let's say uh one agent has the probability of being wrong 10%. Another one has also a probability of being wrong let's say 10%. Which means the chance of both being wrong is around 1%. >> Mhm. >> Right. So uh it is so if you have multi- aent system it's usually um a good and safer way. Then on top of it what we have is uh other mechanics using which we kind of you know uh test it out if this is the answer right let's say there's a question sometimes it lies right LLM's lie and LLMs are unreliable right so we have built the mechanics around it so that let's say given a question does it require writing code yes it did then Did it write the code? Yes. That means the answer must be right. Okay. That kind >> then you you also provide you also provide uh human inputs uh to validate whether the output uh that the that the code it wrote validate that the code it wrote is correct or not. Do you provide human input? >> It runs the code. It runs the code. If the code is wrong, it corrects it. It runs the test cases by itself and then checks whether the outcome is correct. It also checks if the the code it wrote is memory overflowing, it controls that. If the code it wrote is going in infinite loop, it corrects that. So it does all of that. >> Oh, I mean the test cases, right? The test cases you provide it or it generates the test cases. >> It generate. >> Okay. >> Yeah. Because it's human humanly not possible. The only thing we do is we take the feedback from the user saying was it successful? Did you like the result? And if the person says yes, it persists the knowledge it gained by that exercise and it basically if you say no then it tries the second method other method other hypothesis and then if you say yes then it memorize it learns it learns about things from the whole exercise and it keeps on going like that. >> Okay. Okay. So there is there needs to be a human in the loop somewhere to verify >> if you want most of the time it is it does a pretty good job. So initially what we do is that since it's a corporate environment because we cannot fail right. So what we usually do is we take say 20 to 30 use cases for an organization and we basically run say 30 use casea 30 test cases every day to check if the quality is still uh perfect and we keep an eye on that at every >> okay and we run in like multiple ways and observe that if it's giving the good results or not. >> Okay. Okay. Thank you. >> Wonderful questions. Wonderful questions. >> All right. Wonderful. Now I want to show you the real uh case studies that uh our uh you know the our users of Turno and our interns have prepared for FMCS. Though there are hundreds of hundreds of such case studies that have been done by uh people who have been using turno but there are some some things that I want to highlight. Okay. All right. So one question in FMCG is are discounts killing your profit? Many businesses focus solely on increasing sales through aggressive discounting without realizing that the deepest discounts on certain products can erode your profit margins. The challenge is to identify which products and strategies are hurting profitability. Okay. And fix them using datadriven approach like here. So the first is let's say you want to do this analysis that is discount are discounts killing our profit. Okay so essentially both extremes are wrong giving no discounts giving discounts. Okay. And or you can say that uh like what price is good low price or good high price? High price sales will be low. Uh low price sales could be high but you'll make loss. So there's a price discovery. Similarly, there's a discount discovery time, right? So what you can do is let's say you have the data history of data that discounts on every product. So what you can do is find most and least first step would be probably to find most and least profitable product in categories and then visual visualize the relationship between discounts and profits to discover the pattern such as subcategories with negative profits driven by excessive discounting and so on. And then you can visualize it the the profit map by reason and analyze customer segments to find if certain regions or customer types disproportionately buy loss making discounted item. Okay. So uh then is uh uh uh then the objective would be to uh build a smarter discount plan generate action enable datadriven recommendations and so on. So here let me just show you in action. So this is the this is the report that is prepared by by um a human data scientist using turnup. Okay. So first give me the summary of the data set also visualize the distribution of sales profit and discount. So it says that uh sales were this much was the mean. This was the standard deviation. This was minimum. This was maximum. Profit was uh this much. Okay. Mean profit was 28. Standard deviation was this. Mean uh minimum was this. And so you can see that his standard deviation is just humongous. Right? The standard deviation is humongous. Okay. and and uh right now uh basically the the here you can see the discount the I mean this the discount mean is this much standard deviation is this much minimum is this much maximum is this much next is yeah so so what you see is the the the distribution of sales data and then distribution of profit data And now we go further. We say that visualize sales versus profit by category and subcategory. Okay, it goes around and then basically does the job. Okay, so it says that there's a sales data, there's a profit and there sales and you can see that basically here the sales increase in the sales is also increasing profit in most of the cases. Okay. Now, right. Okay. Here, show me the top 10 most profitable and least profitable product along with the uh percentages of um profit and loss. So here these are let's say this is based on certain data from a FMCG company and then these are the top 10 products with which are which are most profitable. Okay. And it has also given the the the profit profit margin and so on. Okay. Okay. And then next is this one. Top 10 least profitable products are basically this one, this one, this one and so on. Okay. And these are the losses that we have made on that. Okay. Then for each subcategory plot their average discount on one axis and their profit on another. Highlight the subcategory with negative profit. Okay. So here so these are the ones that are negative profit and these are the ones that we are there. Okay. First next is okay. So average profit margin would be um this is average profit margin and then basically it's it's a the profit margin when you put the discount here you make the loss. Okay. Now visualize the map of sales and profit across the across the United States. So it has done. Okay. Moving ahead here. Do high value customers buy different categories compared to low value? So that's another analysis. So say for example in furnitureures furniture the these right these are low values these are high value right the customers basically which one do they buy more. So you can see that in case of technology people buy the high value more as compared to the low value. Okay. So that's a very good insight here. Okay. And right recommend which product should get discounts and which should not. Okay. So it has given the the suggestions that give this this is one one product that you should give this discount on. This is another. This is another. It has given us all right. Avoid discount for these products. Okay. Similarly, yeah. And so on. So you can see very clearly that summarize these findings. We are asking it to summarize the finding. So focus area for profit growth. It has figured out and then it has right categories. So technology is worth investing in office supplies and but it has a dilute margin. Okay. And so on. So you can see that overall profit margin hovers around this much profit. So it's kind of a summary of the entire analysis that it has prepared and yeah it has also prepared these dashboards for summary. Okay. So you can you can see this uh very easily uh in in uh using terminal. This is the link to the history prepared of the same conversation that that Nikita had and make sense right everyone you can take a look at this we'll be sharing all these findings and results. So in interno you can once you're done with uh let's say analysis if you're done with analysis let's say here here I'm done with analysis I can click on share and share with uh you know everyone okay so all right let's go back where we were right next insights are from Koka Koka cocoa data Coco data set is essentially um ch it's a data set from chocolate makers and they they want a clarity on which factors drive higher product ratings and which regions bean type or cocoa percentages represent untapped market potential. So here basically as as part of this analysis that I'm going to walk you through is basically a systematic datadriven decisions around the product. That's the core purpose. Okay. So you can take a look at this. So the steps are pretty simple. Understand the data quality and structure. That will be the first step. Second would be discover what drives ratings. Then we will basically once we have done that we do the anal we analyze market trends and gaps and then we provide datadriven recommendations. So this is another uh such example where basically a detailed analysis of the product with respect to geography and and how it works and finally it basically you must take a look at it read through it. This is pretty useful to know and and it figures out that this right this was the conclusion. So the this was the data set with nine columns these many rows no brand missing key fields were cocoa percent and ratings bean type bean uh origin company and company location and those were the few important uh features. Then key findings were that that overall overall Pearson coefficient which shows that if you like keep increasing like the overall there is no direct correlation between the the cocoa percent versus the sales. Okay. The best one best consumer uh preference is 70 to 79% cocoa brand. Right? n if you increase increase the cocoa percent from 79% then also the customer uh wouldn't like it and if you if you basically if you decrease the cocoa percent below 70 then also people don't like it that's the whole conclusion about that report now then these are the other parts and so on all right give me list of all the prompts that I have given in the conversation So it basically gives that and then right you can ask it to prepare all sorts of things that uh show me colored clean organized plot. This is called by the way this is called what do you call this? Uh there is a name for this plot. I did never knew that. Okay it automatically came up with this kind of a nice chart that this will show a better results. Do you see that? Okay. Okay. You can keep on drilling down. Right. And now I do not know how do I go back. Okay. Oh, here. All right. So, you can take a look at that. And that was the second case study. Third case study is similar. Uh, it's about customer turn and happiness analysis. So you have to identify customers at the risk of churn, understand key warning signs of dissatisfaction, explore demographics and segment the patterns. So you can do that simply by using turno basically right aggregate customer metrics figure out what's important what's not then flag and trace customers then analyze key warning signs build customer health score and segment an action plan. So that's what is here. Okay. So this this case study basically talks about that. Okay. So it has prepared some score and you can take a look at that. All right. Yeah. Okay. So, it does that detailed analysis. You can take a look at that. Okay. And yes, it has done the summary for us here. Okay. So it has basically uh you know given the detailed plan on how to how to minimize the customer churn. Okay. The next is forecasting sales right you want to do the basically sales prediction at certain place and it gathers the data cleans it does the sales pattern analysis then forecast the total sales and then it does the granular forecasting by by category and region. So this is the example whereby it takes some time but it would give you the results of forecasting sales. Okay. So these are these are something that something that people have not you know dreamed of that this can be done in a day or two something that used to take years or months. Okay. So that's what Turno does, right? Many of the manufacturing industry or FMCG and so on, they're finding it fairly difficult to hire data scientists and even if there is a data scientist in the such company, they're just too busy. There are too many things to do and they're not able to finish uh the work. So essentially uh this leads to a lot of uh you know missed opportunities in in FMCG. Okay, that's pretty much. So thank you everyone. Uh feel free to ask questions. I'm open to questions. uh if you can hear me. >> Yeah, I can hear you. >> Uh first of all, thanks uh for the session. Uh my question is uh simple. Uh the beginning of the session you mentioned uh that you know the customers can u uh host or install the turno in their cloud environment. If uh if they choose to install outside their cloud environment what are the kind of cons do they have to face? >> Um not much. the the only thing they have to do is when connecting the database they have to allow our IP addresses to connect to the database other than that the basically other than that there's no there's no difficulty at all >> and that has to be uh any public IP uh private IP >> has to be public IP because we won't be able to access it otherwise >> okay all right thank you >> that's the that's the only downside and yeah Adding on to this s uh we only want the read only access on your database so it's not that much harmful for your enterprise right >> no it could be Rohan it could be like it's a private data at the end of day >> no >> but yeah I understood uh the the answer from the other gentleman also uh but at the at the end of the day I'm looking from a cyber security perspective Yeah. >> So from our perspective we we have like the turno the first and foremost thing that was designed for turno was something called SQL sealed that was designed specifically for protecting the databases and like we use that while accessing your database. to use that so that no matter how how much LLM hallucinate it will not let your database be impacted okay in any way that is the most important part okay so we basically uh there were uh you know there were systems whereby you could protect the computer from executing a program like right the way antivir viruses used to be tested was inside that virtual machine. So those are called Docker and containers those things are very popular in software domain but uh SQL shield the database protection is something that people have not built. So we built that ground up. >> Excellent. Thank you. >> Wonderful. Any any questions for me everyone? Uh if there are no questions, I would like to have one more. >> Go ahead. >> Okay. uh what are the kind of LLMs you are using and uh uh you know especially the insights of the turno behind the behind the screens uh in terms of uh open AI uh or you have your own um LLM's uh and your server how how is it working actually >> uh good question we don't we don't have our LLM model we are using we we are we we can use either Gemini or OpenAI or DeepSeek or any other model which provides the reasoning capability. Okay, we we we prefer the models which have uh ability to reason. The ones that do not have ability to reason usually they end up uh making too many mistakes. Okay. And we do not want that. So yeah we are right now on the turnoai do turno.ai AI uh the website that we have there we are using open AAI but for some customer people are using Gemini some are using mostly open AI >> so sep what you're trying to say is so basically so all the readym made models which are available in the market be it like open AI or geminy or deepseek or anything or models which are readily available so so we we are using those uh readyment models for our activity right >> of course Yeah. >> Okay. So that mean there is no LLM that which you which you people are preparing or anything that uh it is directly which has been designed by open AAI and and so on. So that you are just using it customizing it and you using on on its own. >> Correct. Correct. >> Okay. Okay. Okay. Correct. >> We have done a lot of engineering and machine learning around it. >> Okay. Yes. Got it. Got it. Got it. Got it. Got it. Okay. because there's a lot of lot of work that needs to be done to ensure security, accuracy and you know the optimum way of presentation and so on. So there's a lot of work that needs to be done there. That's where we are doing that. So Sep any any any chance that that that you guys are like basically like curious to develop your own models >> instead of depending on like open air or anything. It doesn't make sense to develop our own model, right? Let them develop the model, right? Let there's a fair amount of competition between various companies. >> So, it's better to just use that. There are companies which are like Olama, they're trying to improve the local models. If that turns out to be good enough tomorrow, we'll use that. >> Right. >> Okay. >> The main value basically taking the LLM. LLM is like the fundamental like the way we had the computer. Somebody needs to make it such that it is usable, right? Um and and we are one of those people who are making it usable. It can learn with people. It like people learn to collaborate with it and uh like how can like how can we make sure that it's secure and it's like very optimum and all of that. So we are primarily focused there and uh yeah we designing the semantic systems and so on because the LLM design is u is a different uh game al together and the thing that we are doing is a different game together. So our focus is very very very centric towards building the data scientist. So it is just kind of a kind of a kind of a plug and play uh that you use it and customize it and then then start uh using it. >> Exactly. Exactly. Exactly. Okay. >> We'll be so far organizations need to install it but soon we'll be launching a desktop version where you can download it and use it. >> All right. All right. All right. All right. Sep I was just wondering like uh whether Tano was uh Tano is able to uh do some kind of forecasting uh based on >> uh time series specifically time series based problems. >> Yeah. Yeah. It it will figure out whether it needs to do time series or other things. >> Okay. >> Yeah. Yeah. Yeah. Thank you. You can just connect your data say that give me the forecasting it'll do all the time series analysis and everything. >> Uh Sanjep will it also be able to handle the scenarios like joining production attrition prediction kind of thing. Uh >> yeah this was the use case right here this was the use case here. >> Uh sorry actually I have joined a little bit late. I mean I will just go through the recording but SEP one thing I want to check uh how can we ensure the data security I mean uh is it something called uh will the data will be somewhere like gone out of our I mean if I have to implement this for any enterprise uh systems right I mean so how the security will be handled here >> security will see the it keeps the data with your organization it doesn't let the data go out of your organization it is not uploading the data to any LLM. It is only sending them uh only the bits of information such as this is my table name, this is my column name, tell me uh right and then it generates the code and then it runs it observes and and does it again and again. >> Yeah. >> So your data stays in your organization data is not getting uploaded on on on any LLM or anywhere. >> Okay. Okay. Got it. So basically I mean uh if we have to create any models kind of thing right it will take some uh months kind of thing. I mean if you have to fine-tune the model for accuracy and all. >> So now if I adopt to this tero can I can I speeden up my process of something called let's assume my model is not really doing good. So is it something called it will automatically able to identify the futures required to increase the model. So will it work in that way? >> Yes it'll try. It'll try it'll see that okay this model is not doing good. So it'll give you suggestion that you should include this feature that feature also it'll drop some of the features to improve the performance. >> Okay. >> Okay. >> So we can try it I mean on any of the things do we have any trial version kind of thing? >> You can just go to the website sign up. You go to website sign up put your data ask questions. >> Sure. Okay. >> It's pretty simple pretty straightforward right? We can just uh come come to tero right and it's pretty straightforward here. Okay. like uh here you can just say new chat and then you can okay you just ask questions attach your data and that's it right you can you can give it a shot >> okay sure thanks All right. Thank you everyone. Have a good day. Wonderful to have all of you in this session and it was really good set of questions uh that you guys asked. Thank you. >> Thank you. >> Thanks a lot. Banks.