Large Language Models Bootcamp- Information Session

Key Takeaways

started in just one minute I see the webinar has started already yeah we see that um I will just give it one more minute uh so we start at 11:00 a.m. Pacific time okay so so welcome uh to the information session for the boot camp everyone my name is Raja ibal uh I'm the chief data scientist and one of the lead instructors at data science Tojo I'm I'm going to walk you through um the boot camp um the boot camp curriculum any I'm happy to answer any questions uh and and so on so I'll go ahead and get started um we have been around for quite some uh time uh we are one of the oldest players in uh in um in data science in machine learning and analytics upscaling uh a lot of graduates lot of companies on our portfolio um pretty much any company that matters on the planet they have been someone from that company has been trained by one of our um uh in one of our trainings um so I will jump right in uh how did the boot camp actually start uh so we'll talk about the boot camp um itself and then we'll see uh you know what is the curriculum and so on so so when we get started when we look at chat G BPT and Bard and Claude uh and we we run these very basic scenarios we are talking about uh you know content generation create a social media post create a um uh uh create um maybe an essay that you're writing for U uh for your one of your courses you perhaps you're um uh perhaps you're writing some code getting some SQL help um there is a lot of um uh I would say misunderstanding or Illusion of understanding that these uh uh these uh tools create uh when you when you go ahead and try to go and Implement an Enterprise application using um using GPD 4 or llama or any of the models whether closed source and open source uh it is actually very hard there is issues that uh uh that you may not have foreseen ahead of time for instance uh um none of us actually pays for um uh for open or for chat GPT or Claud um you know there are free peers I should not say none of us but many of us do not pay uh but it does not mean that these tools come um for free um uh when you go to paid subscription you know you pay maybe 10 20 $30 a month depending upon which tool you're using but uh free tools actually make may make make us think that uh you know well it comes for free but the reality of it is that uh when you when you start using these applications at um in an Enterprise uh the costs actually add up very quickly then there are other issues uh uh you know around uh this is your proprietary Enterprise data um are you going to be putting this data um are you going to be giving this data to um uh to chat GPT or Claude or any of once again uh some of those SAS uh software that is available out there there could be regulatory challenges around that you cannot perhaps you cannot share that data with any of these applications then you're talking about uh you know uh um you know the data governance right so what if you know accidentally your some regulated data or some proprietary data your intellectual property or some uh personally identifiable information what if that makes it uh to chat GPT and um well it is not in your control and this is this is why many companies actually they do not actually allow uh uh allow their employees to use um some of these tools um now whether you use open source or close Source model in your own Enterprise application how do you deal with hallucination how do you deal with uh when the knowledge is constantly moving how do you evaluate the response of a large language model uh the prompts are brittle um um sometimes even if you don't change the prompt still your answers can vary um so there are a lot of issues lot of uh issues broadly speaking from uh um the business challenges like uh you know uh data governance and AI governance challenges the cost Rel related challenges there could be uh adoption related challenges there's some ux design related challenges there's um uh correctness of responses challenges you know there is engineering challenges uh building an Enterprise application it is much more than just going to chat GPT and starting to use it so um the boot camp actually will teach you how to build an llm application on Enterprise data end to end and when I say end to end it really means it right so you will learn actually the entire stack of an llm application whether it is a rag application or a fine-tuned application I will show you the culum uh and the labs that we have and the kind of infrastructure that we have set up let me just make sure yes I'm I'm looking at the current screen here um so uh it is a comprehensive curriculum design by practitioners um um we actually are a unique company in the sense that we are not merely a training company anything that we teach we also actually practice so uh what that means is that we have an enter uh we have built uh an Enterprise uh uh software that we license to uh Fortune 500 companies and we have actually big customers you know you're talking about you know 50 plus 100 plus b doll companies that are using the software that we have built uh for precisely this uh this purpose um we cover um um we cover all topics not purely from a technical standpoint so when we talk about uh a particular particular open source Library we are also going to talk about how did we use it in building the software that we have uh and then of course you will see the software that we have built and how do we use how did we use it the business chall challenges the product managers management challenges the engineering challenges the challenges the deployment challenges in the cloud and so on so we actually it is think of it as a much more um uh getting the bigger picture of uh of building llm applications which is beyond just building um you know just uh learning a one Library here or one Library there um it is a very Hands-On boot camp uh you will will be learning uh anytime when you learn one topic you're going to go and learn the Practical Hands-On aspect of it and then there is a comprehensive project uh that you do at the end of the uh of the week five days um eight hours a day uh fairly intense and on the last day you build your own llm application uh the boot camp actually uh we are bringing this boot camp uh with leading names in uh in this uh space vv8 is one of the leading Vector databases in Market neo4j is uh uh leading in terms of knowledge graphs and then now knowledge graphs they um they can be used with in conjunction with Rag and for llm applications Union is U basically deployment of your um your models arise is a leading platform in observability and guard rails and monitoring tracing of your model front part um uh GPU clouds uh or you know we get the uh GPU compute from them Sano Academy we have L Sano who is uh one of the instructors for the boot camp and the Lang chain is uh think of this as a um uh Lang chain is but uh for the lack of better word I mean uh it's an orchestration I mean not quite an orchestration framework I mean it has some orchestration component to it think of this as a library that will give you pretty much most of the functionality that you need to build an llm application I will elaborate on this uh in a moment and then uh you get up to 500 uh US dollars of credit for all software and Cloud uh Services needed so you do not need to bring anything except a browser enabled laptop no need to install any software uh um we have set up our lab in a manner uh that if uh most of in most cases you will be using our own uh learning infrastructure for running all the labs we have all the patches all the libraries all the dependencies we have installed those and uh and then in case of GPU compute uh we have our partners who are actually um uh we have our partners who will be actually uh uh who will be uh sponsoring uh the compute cost for your uh uh for your um U model fine tuning and other exercises I will elaborate on this and we are partners and speakers from leading companies uh in the world who are uh in this space who uh will be speaking and teaching this boot camp um let me keep going here what I I'm going to do here is very quickly I will go to the curriculum um overview what do we cover uh what is the philosophy um uh what does the learning infrastructure look like and uh will go to the Partnerships and testimonials later um so uh think of uh think of this the bigger picture um when you build an L application uh you need uh for building an llm application um you know chat GPT is a working functional uh llm application when you go there well you see uh and um you you ask a question and then you magically get that answer and is it is actually hard to um appreciate what goes on behind the scenes so um what we will actually talk about is what does it take and how would you actually build an llm application from scratch on your own so that is uh what we will be teaching you so one of the components in building an llm application is well the llms and llms uh can be closed source and open source we touch upon on both uh open source and closed Source llms um in terms of um uh Vector databases Vector databases are going to be a key component uh in in this entire ecosystem we'll talk about uh Vector databases uh when you are dealing with Vector databases you need embeddings uh or models that the encoder models the models that actually create these embeddings we'll talk about that then um orchestration a few things can come under under this category uh you have I mean a bunch of python llama index and uh you know other um Lang chain uh type tools uh we cover uh Lang chain in a lot of depth once you deploy an application you have to monitor you have to be able to monitor these applications so we will get into the monitoring and guard rails aspect how do you log things how do you actually apply guard rails on your prompts and also on the responses so um in short in in in summary we have uh we actually um cover the entire stack uh as you would expect uh you know any good comprehensive curriculum uh to do let me actually go and uh uh just see what we have um and very often people ask about uh prerequisites um uh all we expect you to know is uh some basic Python Programming um understanding and we have tutorials to actually get you started um uh even before the boot camp so when you do this boot camp in person or even some people they attend in person some people they are joining remotely and uh you uh um so at the same time some people are sitting in the class other people are remote join uh joining remotely um and that's it right so and this this this is a requirement that can be met very easily uh but if you have never ever done any kind of programming no SQL not you have not done any SQL at all you've not done any basic uh programming ever it could be difficult uh but anyone who has at least programmed and they are in the technology area they can pick up python I think rest of it uh we will actually teach um so this is a technology stack that we use at different points in time in the curriculum uh I will actually switch to our learning platform and I will uh I will actually show you uh what we have so this is our llm models uh large language models boot camp um so when you sign up you get access to this uh learning platform you can see on the left here there is um um different topics so all of these topics are going to be covered I will go through each of them um in different levels of detail uh in some cases I will provide more details and other cases you know I will just mention that this is what we do um so uh when you come in we uh start with a session about two to three hours of session to give you the bigger picture uh you know well uh why do you need a vector database what are embeddings and uh you know what does it mean to even create embeddings uh why a vector database um then why do you need guard rails what is a prompt what is fine-tuning what is rag um uh and what are the risks when you're building an Enterprise application once you have done that we start in within the first three two to three hours we have set the bigger picture and uh bigger context we start to get inside the business of embeddings and when we talk about embeddings let me see if this is the right slide deck not this one actually this one okay let me actually I see that there lot of questions have filed up so let me actually handle the questions maybe I will handle them backward is this boot camp just videos and reading self-paced or live session this is actually a very very live session John this is your question um this is a live and super interactive uh sessions lot of intens discussions that are happening and so on so uh we um uh we actually do um a live sessions some people attend remotely other people they are actually they attend uh training in our uh training facility here at uh data science toob um then the next question here is what is the reason for having multiple Vector DBS and orchestration Etc I mean I don't know I mean every one is trying to grab a market share John uh so uh you know what is the reason to have multiple uh companies launching their own word processors web browser so I mean that's true with anything so different companies they are coming up with different Vector databases uh some companies are AI first Vector database companies uh um you know like vv8 and uh Pine con um and Milas and zillas I mean some of these companies they started as Vector databases other companies they are catching up uh like mongodb postest they are actually they were traditionally no SQL or SQL databases and now they are actually coming up with own their own Vector functionality okay uh let me see uh John I I know you have another question I will come back to it I want to make sure that I have answered other people's question what Vector database will be used in the curriculum uh traditionally um um we use bv8 which is one of the uh leading one of the top Vector databases in industry but uh you have ex we have exercises in our learning platform I will show you in a bit Shar this is your question I will show you in a bit I mean we have uh examples where you can actually the same examples in Microsoft uh um uh Azure search which also offers ve Vector functionality and radius so we uh have in class we will only be covering bb8 in detail but you have uh self-based exercises that you can perform in other uh uh in other um Vector databases other vendors uh what is the chunk size used to store Vector in database I have no idea right it depends upon the application and this is actually the discussion that we actually have in quite a bit of detail when uh we are um uh when we are um you know discussing rag application because the chunk size actually varies uh from application to application and it is well the answer is it depends are there discounts for the boot camp if online only there are actually uh discounts uh there's no exclusive discount for online only uh I'm assuming online uh we don't have a self-paced version so if you're attending remotely the price is about the same but reach out uh and we are happy to uh uh happy to talk uh do you have give access to physical servers which have Nidia gpus for used for training uh models as part of the lab project no I mean we do not give um I wish we had the um money to actually host that many gpus uh or that many machines um uh for you know 20ish people but uh I assure you that the what infrastructure that we give you behind the scenes there's a physical server setting I mean I don't know if you want to see the physical server how it looks like uh but uh in reality the exercises it doesn't matter whether Nu NV Nvidia gpus are physically here or they are physically somewhere else and you're connecting with them so they will be somewhere else they will be in runpod Cloud uh they have given us uh they provide credit for that so runp part gives us credit and then uh we share those and then you connect to those servers that are I assure you they're physically somewhere in some data center but you will be connecting with them remotely let me see where can we find the recording after the session it will be posted um you know uh you are on the I think you will get a link to the recording since you are on the zoom call John so you should be getting a link right after for those watching on LinkedIn this will become a LinkedIn live session right after let me okay the other question so as the output from the boot camp would we have a production ready llm application that we could actually sell in the market no of course not John I mean if that it was if it was that easy uh um no but you will be able to build it right so uh just imagine right so uh it is it is non-trivial Pursuit uh because I don't know whether you know coding or not what's your background uh but uh you will know all the nuances that it will take and you will have a basic working prototype by the end of the boot camp but no I mean be cannot we certainly cannot guarantee that you will have an application that you will be able to sell by the end of the boot camp you know it product development simply does not work like this so that's not going to happen um do you cover performance tuning in the boot camp I don't know what that mean Shar uh we do talk a lot about uh um um will um uh we talk do talk a lot about performance I don't know what you mean by Performance Tuning uh yes we do talk a lot a lot about it right um um will attendees get job after completing the project I mean we are not um so we are not a job placement boot camp um and no I mean there's no guarantee I think uh it is just a boot camp and that's pretty much it right um any prerequisite qualification required for this session I'm from non-coding background considerably if you have basic python skill and if you're interested in uh large language models yes I I think uh you should be able to uh you should be able to you know get by very easily okay Shar I will give a pause because otherwise I mean this is too many questions so I will have to actually give it a pause and I will come back sorry sorry about that right so because I have to actually finish these questions uh finish the session as well uh so let me let me actually go back uh so what we do is we talk about Transformers uh we have L Sano he actually teaches the Transformer architecture talks about you know uh you know anything that you can imagine around Transformers I mean this is a fairly fairly detailed session about uh how do you how do you build actually um um let me actually so Zoom it in so you can see that we talk about uh you know AI sentiment analysis and we later on we talk about the encoder decoder architecture I think my machine is actually taking a bit and uh some time to actually load these slides I don't know what is going on browser may be out of memory but you guys see this um you know it is there uh and then we have practical exercises on you know creating embeddings and um uh and uh you know all the Transformer related exercises let me give you this in detail um you know this session on Vector databases so when we go to Vector databases um let me actually show you what we discuss I will I will just expand on one section there's no way I can I can um I can actually cover every single thing that happens in the boot camp so when we talk about a vector database uh we talk about uh you know the why do we need a vector databases and then what are the different indexing approaches uh uh you know for instance you know you see uh this hnsw index uh uh why do you need Vector compression how do you scale um and then um the really U uh hybrid search U uh Vector search semantic search we talk about all of those topics uh we cover all of these in uh at length you can see Vector searches keyword searches hybrid searches and then in addition to that you need some sort of filtering um your traditional um you know SQL like querying where you are actually of quering based on some facets um then um we are um you know uh I will just scroll through the slides and we explain you know why do you need this kind of index how do you find um this needle in a ha stack in the sense that you're trying to find a single uh document uh or single you're trying to find a match for an embedding uh where there are potentially you know billions of embeddings and how do you actually get that submillisecond response from these billions of uh embeddings out there um and then we talk about you know these layered proximity graphs hnsw uh we talk about you know then slowly the discussion starts going towards uh toward rag uh and so on so once that is um that is done we are going to go into practical exercises uh and then you can see that we have exercises around Vector search and similarity search and then um generative search and product compression and sematic caching and so on so we talk about all of these ideas in theory and then we go ahead and implement this let me click on just one of the labs uh the way this LAB Works is out of all of these six seven Labs that were are so we do all of them click on them one by one and now you can see that you have this uh you know a vv8 server that is already deployed then we create a collection uh you we import the data into a uh into a vv8 vector database then we do a hybrid search we play around with these parameters here we try to understand actually you know how many search results are want uh Alpha equals one you know whether it's a fully hybrid search or fully semantic search or maybe something in between a bit of hybrid bit of semantic search um or um or combination of both um then adding more um adding uh some kind of filter so uh we talk about not just the theory aspect of it you will see all of this in action so that is uh going to be the uh the idea um now um now you can see that we are talking about Advanced rack we have this fundamentals of prompt engineering uh fine-tuning uh when we talk about fine tuning um we actually give you access to um a run everyone gets uh some G credits to uh you know this runp part GPU cloud and then uh we give you llama 2 7 billion 4bit monetized uh model uh you download it from hugging face you fine tune you get a data set you fine-tune it and compare uh the compare the uh the fine-tune model with the model that was has not been fine-tuned uh you play around with different parameters you know more apox less AO know changing different parameters in fine-tuning and uh even before we actually do the Hands-On exercise we first talk about you know what is fine-tuning what is transfer learning what is quantization what is low rank adaptation what is um qora which is quantized low R low rank adaptation so we talk about all of these uh issues we so first we talk about the theory then we talk about the the Practical side of it and we try to understand you know when would you when would you use fine tuning when would you use a rag based architecture um once that is done uh then you are going to uh you're going to um go to Lang chain uh and we have uh in Lang chain you can see that we cover the entire Lang chain stack we have we talk about model IO we talk about retrieval um what is model iio u i mean think of it as Lang chain being a library that um that allows you to create templates of prompts um you can uh you can have you can connect to different types of models you can um you can leverage um maybe something in Amazon Bedrock or Azure or open AI or meta or anthropic um I will show you maybe just one example just to give you a sense uh um you know we we we talk about a lot of different things here so for instance uh different kind of document loaders different kind of uh text Splitters Vector Stores um and so on on uh how do you um how do you control the chunk size how do you connect how do you par a PDF uh and so on you can see that we have uh these examples we are somewhere in this uh you know we are actually reading it up and uh you know just looking at these chunks uh where are we so I I will show you this example here uh we talk about chains uh different kind of chains are possible simple sequential ch chain summarization chains so for instance right so what would it look like we will give you uh we'll give you uh a key uh for U let me just copy and paste the key my my key here so we'll give you a a um and um open AI key and you will be building so while you do this exercise so just Lang chain is about six to8 hours of pure Hands-On work um you know each of the uh that you see on the left model retrievable memory agents uh this is your agentic uh framework Lang chain Lang graph about 6 to8 hours of quite intense work and when you are uh doing the labs so you don't have to worry about installing anything we don't want you to worry about dependencies so in this case let me copy and paste my key here you can see that this is talking about uh some some chains uh so this is an example of a single simple qu quential chain uh and then you run run run and in this case you can see that it is calling an open Ai call uh it is making an open ey call and then um you know this is the router chain um and then once again it is making an open eye call and then running it so um so basically um we are not just covering things in theory there is uh you know we are actually doing doing things in practice now um we talk about agents uh in quite a bit of detail and then we all recently also uh as Lang graph is becoming more popular uh building these agentic uh systems we are also uh we are also actually covering Lang graph in quite a lot of detail I will give you an idea agents and tools uh what is agents and tools you basically have different um uh you have different different um well tools I don't know what to call them uh you know you have a Search tool you have a you have web search tool you have a API calling tool so you are actually now letting llm uh work on different tasks and decide what task actually uh should be performed based on um based on context so in this case I believe this exercise is with the duck Duo Search tool and if you look at this let me see I will have to copy my open I key here so if you look at this here uh this is an example of it will go and selectively um go and search for uh okay you know this is from the last boot camp uh this was before the election so you can see but maybe I can say uh you know I can ask who is the next president of the United States right so and it will go and inqu it will query in real time and it will come back and give me an answer if you ask this question to a standard um uh non- agentic large language models large language models are uh their knowledge is Frozen in time at the time of training uh and those models would not know what is uh you know who's the next president because they are not not aware or they are not familiar with the uh they are not familiar with the um they're not familiar with what is happening uh you know they don't even know what the current date is right so when you need uh this kind of behavior uh that you want to use the language or the semantic capability of a large language model and combine it with uh and combine it with the um you know some other tools um then you use this kind of agentic uh framework and you can see that it was able to pull um that Donald Trump has won uh the 2024 presidential elections and will serve as the president for a second time if you ask this question to a vanilla base llm it will not be able to answer uh let me go I think this just the outline actually actually takes us uh close to an hour so I will try to rush and make sure that we finish on time uh let me go back to the boot camp now uh so what else I think I've mentioned fine tuning okay now you know how to build an application what do you do how do you keep your application safe how do you log how do you monitor um and for that uh we actually have a about four 4ish hour session where we talk about uh uh monitoring and uh logging and and observability uh and uh you know generally call it llm Ops we actually have our partner arise I should have mentioned I mean for Lang chain well Lang chain is a partner vb8 is our Vector databases partner um runp part actually provides us compute for the fine-tuning exercises so I mean they they are our partners so in this case for observability and guard rails arise is the leading platform arise is our partner that actually helps us uh with this aspect of the boot camp and you will learn how to build safe applications in um uh using any large language model uh for any purpose uh we have we uh spend quite a bit of time on evaluation because if you build something without knowing how to evaluate well you don't know how to build it and in this case if you look at this um uh we are using different kinds of evaluation we have different evaluation Labs ragas is uh really for when you build a retrieval augmented uh generation based U uh based uh application ragas is the framework you can see that we once again we have the entire Library we have um you know all the all of these um exercises that are there we cover um some of them for evaluation some of them them in class and then we leave some to your uh you know for your own um later on for your consumption um but you can see that we are covering uh pretty much um pretty much all aspects of uh an llm building application we also have sessions around where we talk about the product management and the product engineering aspect of uh building an application what are the risks and um and uh how what are the challenges when you're building an application at scale we'll talk about uh uh the software architecture uh you know what what what is an ideal architecture uh we have done it I mean we uh we we have like $500 billion dollar companies that have taken dependency on our framework that we have our platform rather that we have built so you know we know what how to build scalable uh llm applications and going back to the question I think someone asked Hey by the end of the week uh will I have an application that he can sell it took us a year and a half and we are barely started selling so uh a real Entre prise application that can you can monetize on maybe you can do definitely you can start Consulting um Consulting work right after but building a product that people will be willing to Enterprises will be willing to pay I mean it is non-rival Pursuit we have like a three dozen people working on it uh for a year and a half and now we are barely barely now we are selling that product right so we it took us a while uh because Enterprise applications they are not easy to build but you will understand what will entail you will be able to evaluate uh um and you will know what will it take but building it no that will not happen during the week so uh I think that but I will take once again I will pause here and then see I see six more messages here so let me see how many projects uh are part of the boot camp it depends what you consider project I mean a project in a project sense there is only one final project that you will be uh building but uh um uh but um you know I mean there are so many exercises and if you consider those then I mean maybe there is a few dozen different things that you will be doing but eventually everything will come together and then they will become a single uh project what is the difference between Rag and fine tuning that is what we will cover in um um in the boot camp um broadly speaking rag is when you combine a large language Model A Foundation model or large language model with uh with a search type infrastructure retrieval augmented Generation Um and then uh fine-tuning is that you then you modify the model itself and then try to use it for your own specific purpose um yes I mean there could be I heard that rag provides find uning functionality yes I mean yes and no um yes we do cover responsible AI uh I think that is what you meant uh um okay so I think uh I have gone through all of the questions here let me uh let me go over wrap up uh my my slide deck here um um so let me actually go I think I have talked about most of these things uh a lot of detail uh frankly uh I'm you may have noticed I mean I'm trying to rush through things because the amount of material that we cover in this boot camp is uh it is just uh incredible right so the uh we proud that I mean we are the only boot camp in uh at the moment in the world I mean depending upon the word boot camp I mean just in a in a short I mean you can call anything a boot camp I mean I've seen like a dayong boot camp as well but I mean this is a 40h hour very intense training and we have seen uh phenomenal uh results uh people have gone back and they've started building um things they've started building things um when they went back and uh they found it to be very very useful very complet very Hands-On one question that comes up is do I have to be a developer no you don't have to be you can be a Founder you can be a CEO you can be a Dev manager you can be a technical product manager um as a matter of fact I would strongly encourage anyone who is doing technical product management manager management for an llm product uh to to learn this skill because uh you cannot there is absolutely no way that you can actually product manage a product that is an llm product uh without knowing all the nuances um there is so many uh what I call the landmines when you're building these applications um you know it is actually um it will help you uh become a very seasoned llm product manager um if or very experienced product manager in the sense that U you it will give the technical underpinnings of llms if you if you go through the boot camp um so this is uh expected instructors uh for the next uh uh for next time uh we have a bunch of different instructors who show up uh from different companies I actually Le I teach about 30 to 40% of the boot camp and uh um and then rest is actually taught by some of our partners and other uh people who attend and here is here's some of our partners who actually think that uh you know we have we do a great job in U in teaching people um and uh here's bunch of logos I mean if you look at this um they are there are this is not by no means this is complete uh list I mean we cannot uh give up we have people traveling from overseas uh you can see some of the like really big uh you know logos trillion dollar companies they are sending people to us for getting trained we' have done uh Enterprise trainings for big companies you know in the US and and uh outside of the US um so I mean we have been this in this space for quite some time uh for a long time and now in llms uh I mean we apparently we are the only training company at the moment that is doing these boot camps for sure uh I expect more to come in and they have been doing this for almost more than a year now um so the next boot camp is happening uh in Seattle if you prefer to join in person December 2nd to 6th uh you can more than welcome to come in of course uh on the in the in the in-person setting the seating and the capacity is limited uh but the same boot camp we have you know the the classroom is set up in a manner that remotely people attend they Dial Dial in and you know they can see everyone in class everyone in class can see them and you know all those multi- viw cameras Etc so we have everything in place uh we have this classroom uh and then it's a very interactive experience even for those uh if you're located uh far away uh you should you should be fine you should be able to actually attend um remotely as well and uh I think that is it uh on my end let me see if there are any questions here any Q&A any more questions uh okay I'm happy to take any questions but uh uh the projects uh Shar we are can you explain the project that we will do you will be building a basic rag application um or I'm putting in guard rails putting in different kind of chains uh in a single application the way this will work is we'll give you boilerplate code uh for a basic llm application with a very basic uh you know web experience so and then you will be just deploying it and then we'll give you exercises hey can you add this chain can you add memory feature can you add PDF uh upload feature or not so um and that's uh that's what we'll do um so it it will be a very basic agent you're starting with a single uh uh single assistant and then that and adding some more agentic Behavior to it so that is what the project is going to be at a very high level okay so um are there any questions about uh anything uh where is the vector database stored um I don't know uh Shar if I can ask answer your question is it an aw ss3 or block storage I uh do not know because uh you know we are using bb8 uh and it varies from uh um it depends upon you know which vendor you are using right and it is definitely not S3 I can tell you this uh uh because the kind of story storage that you need for a vector database it is very different from your S3 or block storage but uh and that is the detail that we will be actually covering during our uh during our uh boot camp um the agenda uh yes someone from the team can share uh on a link if you search for llm boot camp um um the our llm boot camp page has the entire agenda uh of uh you know entire curriculum it is listed there okay um [Music] so thank you so much everyone and uh we would uh uh we are looking forward to seeing at least some of you at the boot camp uh please feel free to reach out where the team has shared uh some links if you would like to talk to uh an uh an advisor or an instructor we are happy to get in a call uh to help you out thanks everyone have a great rest of the day