Large Language Models Bootcamp- Information Session

Key Takeaways

okay and uh we'll get started in a few moments I think we are live now uh so we'll get started so welcome to the information session everyone uh my name is Raja abbal I'm the chief data scientist and one of the lead instructors at data science Dojo um so today's session is about uh our large language models boot camp and we are going to be talking about the general uh Logistics of the boot camp and then also the curriculum of the boot camp and and uh what is the philosophy what are you going to learn how much Theory how much uh how much practice uh what the boot camp is and what it is not so we'll get started right away um so we have been in this space for quite some time one of the oldest and one of the most respected companies in uh machine learning and AI upskilling space um lot of uh a lot of companies have been upscaled by us um um more than 11 12,000 graduates uh basically a long time uh we've been doing this and then uh our Flagship product has been uh the our data science boot camp um and then last year we launched a large language models boot camp uh the first in the world um and as far as we know even until now there's no other boot camp uh you can call it a boot camp but boot camp has a certain connotation of course uh it's a very intense uh uh 40-hour training where you start with U you come with some background in technology some background in coding and you leave with your own working llm application right so not all boot camps are created equal so um uh some of the common use cases as we all know uh I think even you know you don't have to be a llm application developer or a research scientist uh to know that b and Chad GPT and Claud and some of these they are actually absolutely amazing um and sometimes uh actually most of the time they impress us with uh their answers um I can write social media posts I can create uh you know summaries I can write emails I can uh debug code and all of that but as it turns out um U these small uh oneoff cases the consumer cases as I I would call them um they're easy to crack they are easy uh in terms of uh in most cases well if it doesn't work uh there is no business and there's no serious business impact but when you deploy or when you employ these uh when you employ these uh uh when you employ these uh Technologies um in Enterprise um there is uh there's a lot of factors that you need to be mindful of uh there is this token usage and costs right so these uh Technologies these tools they are not free um and as a matter of fact most of the Enterprises most of the product that are out there except for Hardware manufacturers uh no most notably Nvidia most of them are actually losing money because you know Enterprises are still in the mode of discovering use cases so the cost is a consideration you have to architect your application um in in a manner that it is it is cost efficient then there are other uh limitations like context Windows uh when you when you build a product like this uh you have to be mindful that uh you know llms are evolving and their attention span um is not as big and yes I mean the context windows are growing um I mean until very recently it was 4K and then we started seeing 16 and 8K and now we have uh you know GPD 40 and llama 3.1 they have reached 128k context window uh which seems like a lot but for real world enter Enterprise cases how do you handle these context Windows that's a that's a question then the next thing is how do you handle the regulatory challenges are you mindful or are you uh are you aware of the regulatory implications uh how do you handle proprietary data how do you handle um u latencies in inference uh Ai and data governance issues uh when to use open source when not to use open source and uh you know what is the tradeoff um then uh how do you handle hallucinations hallucinations are going to be a part of life uh you when you build an application uh there are going to be hallucinations right so and that's that's an interesting area your knowledge is constantly on in motion it is changing evolving um how do you evaluate your models um prompts are brittle how do you handle the brittleness of prompts there are uh tons of along with these there are tons of other factors so the what we realized while building this product um while building a product and we have paid users for the product uh while we were building this product we thought that it may be a good idea to actually build a uh create a boot camp and teach others how to build uh a product like that so and this is one of the differentiators I think even in the previous wave of data science boot camps anyone who um had some PowerPoint skills and knew a few topics here and there uh without pra practical experience many people they started teaching uh data science uh what I will guarantee you in this case and you will see the the profiles of people uh who are teaching our uh in our large language models boot camp they are practitioners they are not someone who just read one paper there attended one conference there these people spend a substantial chunk of their time I'm one of the instructors I spend 60% of my time on a daily basis in Building Product uh discussions that go beyond just AI discussions that go um into how to optimize application lot of lot of this is actually uh software development how do you deploy um your llm application a scalable manner so uh so this boot camp U is not purely about you know everything AI this is about operationalizing this is about uh making sure that you build a product that is going to make uh your customers is happy this is going to actually uh make more money or cut cost and it is going to you know reduce uh and it it will make things faster for your organization so this is more for uh making sure that you're able to use these um use Ai and generative AI to build something that is Meaningful to your company so um we have self-based courses I will not get into that uh but in uh llm boot camp there are two options there is a live uh course um um you have llm boot camp happening in Seattle we mostly stick to Seattle we have been we have been doing it in other parts of the world as well but for now I mean we are sticking to Seattle for the next few uh at least for the rest of uh 2024 um and then uh you can actually also attend the Seattle boot camp remotely and uh so that's an option uh that is purely possible um now so I will be talking about our 40-hour boot camp it's a 40-hour comprehensive boot camp you start on Monday um coming in as a beginner and you'll leave uh with a working llm application of your own and I will show you everything that we have uh the so basically uh comprehensive curriculum uh pretty much all parts of the stack are covered um most of the main mainstream tools and libraries they are not not only that we cover them they are our partners in delivering this boot camp and I will show you what I mean um and the boot camp is very very Hands-On uh it's it's not that we will just keep talking about you know um let's say Vector databases we are not going to only talk about it right so we are actually going to talk about hybrid search we are going to talk about semantic search we we going to talk about compressing your vectors um we are not only going to hand wave about Lang chain will comp cover it in in its entirety we are not going to cover uh we are not going to cover evaluation only in uh in theory we are going to cover evaluation actually evaluation metrics and actually uh doing lab we are not going to only cover talk about guard rails we will actually show you how you implement guard rails in your model so so it's a it's basically some Theory some practice some Theory some practice and the uh keep doing this uh the entire week um and then there at the end of the week there is a comprehensive project with mentoring and support so that's uh on the last day of the pro last day of the boot camp all of us will uh will give you a starting boilerplate code uh even if you're not a a great coder even even then you should be able to actually uh Implement uh this uh Implement uh uh the the project very well um so what we do is uh during the boot camp of course uh generative Ai and is expensive uh the infrastructure is expensive gpus are expensive uh so the boot camp cost actually includes um the boot camp cost actually includes uh um up to $500 up to uh $500 uh during the boot camp if we you're doing any exercises we'll give you open AI Keys uh so you can consume the tokens uh as much as you like up to $500 of course then um we have a fine tuning that is happening teach will teach you how to fine tune models and the cost associated with that any GPU consumption that is happening uh we will give you that uh we'll cover that cost and then there is uh Jupiter notebook works with hundreds of quote samples and uh even that it is going to be covered um for our uh boot camp we have uh for the Seattle boot camp uh uh that is the next boot camp uh we have had speakers from all of these companies at different points in time um and basically these are leading leading people from industry of course I mean getting on their schedule it is hard so uh to that end this time uh we have uh speakers from and speakers and instructors uh from uh you know different companies um and um you know the these are our partners and speakers from um for this boot camp and I will probably elaborate that uh in U in a few minutes um I think we have talked about this so let me let me go over uh the curriculum and explain to you what the what the curriculum is like let me actually go here yeah this one okay so uh think of this as a uh your typically your llm application might consist of a few of these components you will have a large language models either hosted uh somewhere or you are self-hosting it it could be open source it could be something like open AI already deployed it could be open AI deployed in Azure you can be using something from Amazon Bedrock but at a very high level you have uh some kind of um you have some kind of uh uh llm that you're using but simply an llm a logge language model uh or Foundation model is not enough you also need uh in conjunction uh to to build an LM llm application you need a vector database uh and Vector database I will explain that at a very high level um you know you know you when you're building a rag uh application you retrieval augmented generation application you need to actually be able to search your documents retrieve your documents in a keyword or or um in a keyword uh or hybrid search fashion or semantic search fashion um so you need uh Vector database we uh we use one of the leading Vector databases uh bb8 and I will show you the exercises that we cover uh we use the one the leading Vector databases in Industry uh for our uh for the uh for the boot camp about 4ish hours of session then if you look at this there is this idea of uh you know um semantic caching or llm Cache uh basically the idea is uh if you are repeatedly asking the same question especially if the your answers to questions are not changing there's no point actually going and repeatedly asking the same question right uh um you know um and then the semantic caching the idea is in in your software caching what you do is um anything if you hit the same resource again it is going to serve from cach in this case even if it is semantically similar you know you're phrasing it differently but the question is similar to what what was asked previously you may actually build the semantic cach that's uh the idea then there's logging and llm Ops uh uh monitoring observability of your models that is there guard rails is there and I will explain to you uh how what we do in each of these uh then embedding models uh how do you create embeddings out of your documents and chunks and then how do you go about actually setting everything up how do you actually put all of this together how do you set these applications up and put them together so that you do not uh you know you don't have to um do certain tasks repeatedly again and again so I think I would uh instead of hand hand waving I'm going to actually show you uh what we have and then you will be actually deploying uh if it is a web app you will be deploying this somewhere in um you will be deploying somewhere in your uh in your as a web app in Azure in streamlit in uh hugging face and so on so so let me go over and give you an idea um of what our the curriculum looks like so this is our um The Learning platform that we have uh and in this learning platform you will see the topics that we will be covering um uh in detail so we'll start with embeddings um you know how do you actually convert uh piece of text into or for that matter anything else right uh video or uh images how do you convert um something into form of a numeric representation how do you convert a piece of text into numeric representation so we start with historical uh very quickly very quick overview of uh uh what embeddings are have been historically how we created text encodings one hot encoding um to uh count past encoding to tfidf to um uh word to and BT and and so on so so and then slowly you'll see how we started evolving from you know just keyword based embeddings to semantic embeddings slowly and then uh we go to attention mechanism U you know if you are here most likely you have heard of if you're in this call or watching this webinar you probably have heard of why attention mechanism and why Transformers are important um this session Lewis will be teaching it Lano will be teaching this this particular session so well you know we go into the details of um all of the attention mechanism Etc and then we have Hands-On exercises then we move on to Vector databases now you know how to create embeddings out of your text how do you actually go about um uh and uh store these embeddings in a vector database in the in an efficient manner I can uh actually I cannot of course go through every single uh piece of the or every single bit of the curriculum I will show you you can see that there's a bunch of different exercises but in if you look at this here if I come here and then I look at it so we start with uh we have a very detailed session U on Vector databases Zen he's amazing I mean he teaches uh this uh um he takes about two to two and a half hours just going through what are vector databases so we talk about uh Vector database um uh and uh um we talk about Vector databases we see how Vector DBS work how are vector databas is different from your um your regular uh SQL databases um then um you know we talk about yeah you know how how are they different if you look at this you know there's a lot that is going on of course I cannot cover every single piece every single bit here but you can see we talk we'll talk about Vector search keyo search Hybrid search uh then we will be talking about you know what are some of the ways to optimize your uh you know indexing um you know hnsw uh layered approximation graphs and uh layered proximity graphs and so on so we cover Vector databases and then once you have done all of this uh we don't just settle for Theory then we go and do start with the lab on Vector search right so if you look at this the labs are in integrated I will click on this and the labs lab opens up um this is a dedicated storage and compute that all of you get as part of the boot camp registration we bear the cost of the infrastructure uh and these labs are actually you have these for uh one year um up to one year uh uh after the training uh so this particular piece right uh so you have a deployed VV server you can see that uh you know you will go you will create your own embeddings you will store those embeddings you will perform Vector search uh then what you will go to hybrid search and you will see um how do you do sematic search how do you do hybrid search and you can see I can keep going uh if you look at this I can run run run and keep going uh and then you can see generative search multi-tenancy and compression all of these are going to be covered we'll go line by line and um and and discuss and one common question that people ask is um do I need to be a coder to attend this boot camp I do not think so uh I don't think so because we have had people who were not primarily coders and actually if you are a founder or CTO working on a an idea or a technical product manager or a manager of a de team team that is building products or someone who's actually overseeing some vendors who are working on some uh llm application building some llm applications for you uh I cannot emphasize enough on how important it is for you to be technical because um um llm applications are uh very much I would not say very um completely unlike any other anything else but this this is and llm applications gener products are um that the category of products if you're project manager if you're product manager um whoever you are even if you're not writing code you must understand some of the technical details there is no other work around right so you cannot be a great product manager you cannot be a great architect you cannot be a good um Dev manager if you do not understand the tradeoffs so it is going to be a mess uh if you do not understand this and in terms of coding uh if you have some coding background even if you've not done coding in Python we have tutorials for you but if you have some coding backgrounds you can write some basic code um you should be able to follow because you don't have to write code from scratch well if you like to write code you can always modify these Labs save them make a copy and modify them but if you don't like writing code but you still want to get de you know just get get your feet fet I mean reasonably uh competent in on the technical side of it for instance right so if you understood what uh you know you went through this what is a hybrid search in this presentation but if you don't see hybrid search or if you we talked about uh you know uh compression of vectors and we don't show you how vectors are compressed uh you know there's one thing doing it in theory one thing is that is doing it in practice so I would strongly strongly encourage I mean if uh and these are not the things that uh you can learn um just by attending talks uh I mean I'm I'm I have a firm conviction that you cannot learn just by attending talks you have to at least go through some of this uh some of these code samples at least to understand wrap your head around it uh if you are involved in product decisions you have to understand what are the trade-offs well this is Vector databases very detailed session about five wish hours that we spend on Vector databases then we move on to uh retrieval augmented generation the first thing that we cover is Lang chain and in Lang chain if you look at this same thing model iio retrieval chains memory agents um I will I can only show you a few things here right so so for instance how do you create templates out of your um out of uh uh how do you introduce templates in your application right what are templates um how do you handle the situation that your prompts are going to be variable um you know the user will input something and you're going to uh um you're going to actually uh plug in the variable component of the prompt and generate Pro prompts on the Fly based on template how do you do it then how do you um use different kind of chat models how do you uh give it um give your model um uh different kind of examples um we call them few short examples how do you actually go and uh use different models uh different uh uh you know the you're not always going to be using open AI you're not always going to be using Amazon Bedrock or Azure um so Lang chain actually provides all of these U uh this is a framework that offers you um a lot of flexibility so um you know one day you're using one model another day you are using different model so you don't have to go and change your uh other pieces code um because something changed right so it's a it's a and we we go into Lang chain we spend about six hours just on Lang chain maybe more so we will cover uh model IO we'll cover retrieval think of retrieval as connecting it to different data sources your Dropbox your SharePoint your um uh WebMD your open uh some open uh apis uh connecting to SQL Server snowflake you know Azure synapse um AWS Amazon S3 buckets and how do you parse different kind of how you parse different kind of data formats Json PDF so we actually cover all of that uh you know all the labs uh and much more is there and uh the nice thing is you have everything already installed right so when we take click on for instance document loaders all the ne necessary dependencies they are already installed our Engineers have actually got uh you know they have uh gotten everything ready for you so you don't have to waste time on uh you don't waste time on uh you know installing packages when you would you could be actually building application okay so this is there um and you can see that you know text spitter different kind of vector stores and um markdown and so on and if you want to write your customer code you're more than welcome to because you have dedicated storage and dedicated compute under your name in this uh already um then uh we also talk about chains how do you actually uh um break down your prompts and break down your instructions to the model as a series of um inference calls as opposed to a single giant uh inference call um in this case um you know if I go and look at this when I go to um you know there is this thing called a simple sequential chain and when I look at it simple sequential chain is uh um you're breaking down the instructions into two uh two instructions in this case you know there is this the first part of the chain is your job is to come up with the classic dish from the area that the user suggests and the second part is now that you have found out the the a dish uh given give uh given that dish give a re and then you can actually you can see that it first runs the first llm call then it runs the second llm call and uh maybe I can just run it and show you and then if you look at this uh you will need an open AI key which will be given to you by us let me copy and paste this here I have the key here I hope this works okay I will go here run and run and let me actually change this to maybe know uh London right so I will just change to London and if I run this chain and now if you look at this and this will tell me uh you know the first thing I propose fish and chips is a classic dish from London it consists of deep fried batter fish and Then followed by um uh you know uh it will give me a recipe for fish and chips and now if you look at this and if I change it to something else uh Seattle right so this is where I am based if I run it for Seattle and now you can see that it is a combination of a template and uh a chain so what we do is we slowly incrementally build on top of uh the concepts right so um if you look at this a lot of times many of us are motivated we want to learn but we don't know where to start we don't know what to do first and what to do later what exercises would make sense so um summarization chains um once again uh when you have uh when you have a big piece of text so you can be using summarization chain uh there is a chain called map reduce you can be using map reduce chain and there are other chains like stuffed chain and refin chain and then we talk about all of those what those are and then we'll see how we use St we talk about agents you can see that we have exercises for agentic behavior um I will we talk about why do you need agentic behavior and if you look at this uh you know agents and tools uh working with agents uh I will click on just randomly one of them these this is a lot of work by the way the boot camp is intense uh it is um you know sometimes people would say yeah I mean maybe you know it's uh it's not easy by any stretch of imagination but you know as people call it I mean it it is brutal but in a good way right so because well anything non-trivial in life you have to spend time to learn it but we are very confident once you actually go through the boot camp you will be actually ready to build applications or maybe not build applications yeah maybe uh whatever goal you came up with I want I'm a product manager I want to be a technical product manager for LM applications great I mean you can do it I'm a CTO guiding some product uh in uh in general for uh that involves llms yes I mean you should be able to so it gives a complete overview of the ecosystem no corners cut right so you know yes I mean I know that there are courses available one hour course in Lang chain here one hour course in llama index there one hour course in you know Vector databases or two hour course in Vector databases but you know it is a comprehensive course it's not actually maybe a few things here and few things there it is as comprehensive as we could make it now talked about Vector databases talked about embeddings and uh and transer Transformers we've talked about Lang chain then we actually talk about now that uh you already understand uh you have a fair amount of uh context fair amount of uh background so we then get into um you know uh things like uh uh rag and what are the challenges in rag applications then we get into um get into fine-tuning how do you fine-tune a a model so and when we talk about fine tuning once again we're going to talk about the theory quite a bit of theory we'll get into the theory part of fine fine tuning we'll talk about Wi-Fi tuning uh we'll talk about quantization we talk about uh low rank adaptation uh and a high level understanding some math but not too much math of course I mean the math for Laura and quantization can be actually tricky but enough math uh not I I don't like completely sking math but also this is not a theory class this is a practitioner class and once you have done it uh we once you have understood the theory part of it uh then we go and give you a quote sample to fine-tune a llama 2 model um um uh we will um share a a code that actually um there uh that I'm sorry I I saw this comment and it just got sidetracked so I will I will be answering questions I saw a question pop up and I got distracted so we'll be sharing a notebook with you uh that will allow you to um um fine-tune a forbit quantized llama 3 model or um Lama 2 model I'm sorry we are not at llama 3 yet maybe this time we will be able to make changes to our Labs uh but we do a llama 2 model uh we look at U the base model we look at the fine tune model We compare their performance and then we try to understand um you know what does fine tuning entail uh you get the GPU compute right so we give you a GPU cluster our partner uh runpod uh they they give GPU credit to all the attendees so we will give you give you the GPU credits for uh for using uh for fine tuning the the Llama model during the boot camp um and then we'll leave you with more exercises if you want to fine tune U an open AI model uh we'll give you the tutorials how to do it but in general we you are enabled and you have you have seen um this whole um whole cycle of going through fine tuning for a llama model um in addition to that um um Let me let me actually handle some of the questions because I mean the questions are actually coming in uh let me see okay I can't see my face properly is anyone else having issues I'm not sure because I can see myself in Zoom I'm not sure is am I uh am I visible audible properly okay I would assume okay thank you uh I'm not sure why that would be the case Okay do you offer uh discounts for the boot camp yes we do uh please uh set up a call I think uh of the core staff has been posting links to setting up a call I'm happy to get on a call or someone else will is happy to get on a call to actually walk you through the curriculum and then you please reach out and then definitely we can um we can actually go and offer the discounts and uh by byid from YouTube so there's a question uh please share the notebook file I mean there was a lot of notebook files uh that we had here uh so I'm not sure which one is referring to uh which one which notebook you're referring to and a lot of these are basic basically open source code samples I mean I'm not going to you know um some of them we have written some of them are an open source if you're referring to the L chain uh many of them are actually from open source Lang chain what we have done is uh we have actually taken it and organized them and combine them with you know basically when you should be doing what uh so a lot of this is open source if you look it up you should be able to find it um you have a question for the first from the first slide can you open the slide again which is having the architecture so asan if you allow me to finish the presentation then I will definitely go back uh because you know I would I would like to cover the topic I hope this is okay uh with you uh Aran can you tell what are the prerequisites U not much if you can write code in Python and if you're in general interested in uh or if you can understand Python and if your general interest in building a large language model application you should be able to uh do a you should be able to carry out you know all the activ exercises in the boot camp there is another question what would you say is the percentage of theory versus practice time for the boot camp and I would ask you flip your back the question I mean do you want more Theory or more practice I would say it is the right balance uh Lang chain is 80% practice and 20% Theory um fine tuning is 50/50 uh rag um nuances of rag uh it's a two-hour session purely Theory so I would say maybe around 50/50 but it depends upon the session right so because as a practitioner I will tell you there is no such thing I want more practice or more theory in some cases you need more theory in some cases uh uh it is practice okay so that's that's the general idea that's my honest answer I don't know what I can what else I can tell you okay um but uh I mean rest assured that you know everyone who teaches at the boot camp they are actually practitioners and you know they will do whatever is right thing to do right so sometimes we'll have intense discussions U on certain aspects and we can go off script as well right so because uh and also we have a product uh we actually uh show you the product how how it is built and then we discuss uh how would you build if you had to build this you know it's it's it's a combination of um presentations discussions and uh Labs so it's and project of course uh where was I domain specific models this is one of my favorite sessions uh uh from uh so there is this whole idea whole wave of new models which is uh domain uh small language models or domain specific model so caric from symol uh um Syle um AI they are um they have a conversational model called nebula so kic actually walks us through is one of the greatest best instructors my favorite instructors he he teaches he he actually explains why do you need small language model why do you need domain specific model what are the challenges in domain specific models and then we uh we actually go through that um then uh we are uh we talk about uh we our newest partner is arise arise is uh uh uh once you start let me before I talk about arise tell um maybe I can tell you why why why the need uh to have have a partner like them right so just like your devops just like your monitoring and logging just like you know having some kind of security large language models also need all of this um uh large language models also need um all of these aspects so arise uh is a company that does uh llm Ops and monitoring and uh observability um you know how can you um get feedback set up feedback loops on uh on your llm applications how can you cluster the prompts excuse me how can you put guardrails on uh how can you put guard rails on your models all of that so uh we will have a discussion on llm monitoring or presentation and followed by Hands-On exercise of course whenever I say something is happening very likely it has a Hands-On component right unless otherwise stated so well um how do you log your prompts how do you log your feedback how do you log how do you cluster your prompts uh and to understand which prompts are doing well which one promps are not doing well how do you monitor your model um how do you put guardrails that your model your model does not give an answer your llm application does not give an answer that's uh uh that shouldn't be there and how do you make sure that uh you know how do you block prompts uh that shouldn't be there so it's all around you know um more around dipl I should say operationalization right so that's a that's a better word I was missing that so how do you operationalize a model it it is going from uh you know monitoring and logging and uh feedback collection and putting setting up guard rails so uh we will have a we have a session on that uh llm observability observability and evaluation uh and evaluation by the way right so um in in a traditional machine learning sense if you're building a fraud prediction model a fraud prediction model well uh if uh the original label was fraud and you predicted it is not fraud well it's a wrong prediction but U if I if I says uh said uh I I am hungry or uh I really want to eat something right so both of them semantically um they are related but um syntactically in terms of keywords they are not related so how do you evaluate large language models and there is a lot I mean this is a very simple natural language example that I'm giving you uh but there are much more nuanced uh um uh aspects of evaluation and we cover all of them uh will'll talk about you know context recall context Precision answer relevancy and all of that so so we will be talking about um all aspects of uh uh of evaluation um not just the okay um then we have a talk uh a session uh it's not a talk so um we have a session on uh you know deployment and um uh deployment of a model and we will actually uh how do you this is through our partner Sage Elliot from Union he's going to actually show us how do you take a model um and find unit and deploy it and start you know set up set up uh inference uh on that on your own um and uh and on the last day we are going to have uh we're going to have u a project uh last second half of the last dat uh we will give you a boilerplate code uh we'll give you you know you will be deploying uh and you uh the boilerplate code will come uh in a VM everyone gets their own VM you go log in into into VM and then you deploy a web application uh that has an llm application uh so you deploy it uh we give you the API Keys Etc everything that is needed um and then uh you go and deploy it uh you push it into GitHub GitHub goes to you know streamlet cloud and then you have a URL and then we give you exercises can you add this chain can you add this crail can you do this can you do that so basically we tell you the EV we make you familiar with the evolution of uh an LM llm application so you understand how this whole thing works uh what else I'm sure I'm I'm sure I'm forgetting something here but this was a very very very highspeed introduction to uh the curriculum one more thing right so that is uh I'm I think I briefly alluded to that there is a two hour two two to three hour session where we actually talk about some of the actual practical challenges in implementing rag applications not found in textbooks there are no textbooks for some of the real issues uh those are uh you know actual issues that uh we ran into it's really not from any textbook it is from our experience so so how what kind of challenges you can expect or anticipate to run into when you're building a rag application okay let me see I will uh we are all we are actually almost out of time um uh would the sessions be recorded uh so that one can go through them again in case they did not um we usually record the sessions they they are not properly edited and all of that because I mean that's going to be very time consuming but yes we do record the sessions but it's not something that we guarantee that every session because sometimes you know we don't have a full-time video grapher sitting there so for the most part yes sessions are going to be recorded I mean are they going to be perfectly capturing everything at the beginning at the end I mean sometimes you may have something extra and all of that so as long as that is clear it's not going to be very finished self-based learning type recording but if you want to refresh your memory yes that is the case uh prerequisites and some someone asked this question yes I mean we have tutorials even if you don't know Python Programming uh we will we have tutorials to uh get you ready for the boot camp and most importantly you know you should know why you are doing this because I mean if you do not know why you are attending the boot camp is it because uh it is something cool I want to go and learn just have an idea um what you are planning to do uh this is the technology stack uh that we broadly cover in our boot camp um um different you will run into a bunch of these Technologies at different junctures different points in time in our uh and during the entirety of the fot Camp so I will let you see um and maybe if you this should give you an idea uh and this is everything that I've already mentioned uh lot of material right so you know it's uh it's a lot of material I mean if uh in most cases I think for many people if they are even if they are very disciplined it will take a to actually finish all of this cont I don't know how long I cannot put a time frame on this but it is non-trivial amount of work that we actually cover so if you look at this this is all what we talk about during llm evaluation uh multi-agent workflows right so we have multi-agent workflows that we talk about building a lb application this is uh our instructors and speakers uh uh um that uh who will be talking and we have had more speakers in the past these are confirmed very likely we'll have one more two more speakers actually but depending on for confirmation uh one is from arise uh I mean they are they are coming but we don't know who's going to be the speaker yet uh because it's based on availability and one more speaker um who has been giving a talk on building in real world application um that uh that's a very good talk uh you know more from the business side of it but her availability uh we are trying to find out uh so these are our I mean go check out right so we are a fairly uh well-known name uh and our testimonials the great thing is that they're not from any John Doe and Jane do these are real people right so go check out right so talk to these people uh these people and these are the companies um who have attended uh these are there are some big names I don't think this list is complete uh we have had people actually that travel as far as from Australia to attend this boot camp so you know you can rest assured that uh you know we know what we are doing uh the next boot camp is going to be in Seattle uh from August 19th to 23rd and uh um the same boot camp will be uh live casting um uh in if you cannot travel uh because of budgetary constraints or any other uh person constraints uh you can attend the boot camp online exactly the same experience we have you know the classroom set up that you can see everyone in the classroom everyone in the classroom can see you uh and in the last we we started this uh last time in our boot camp in June end of June or early July I'm forgetting uh and and then uh the boot camp uh that that we did there the our online attendees were actually very very engaged um having said that of course in person is uh in person online is online I will let you decide same curriculum all the labs everything is streamlined we here to help you um and um it's the same um 8:30 9 to 9 to5 U every day um and let me see what else is there thank you uh I'm happy to take any questions if you have any questions please um let me know and I'm happy to answer any other question that you may have oh okay I think there was someone who wanted to see one of the slides uh this was one of you let me actually go back and there was this SL and there was a question so yes uh please go ahead what was the question uh also you mentioned that there are three to four options of apps to build during the last day would you be providing sample answers yes we give you code samples um um we actually will give you in the class we will be doing only one app because I mean it's a it's the uh it's more it has more to do with pedagogy then really I mean the keeping the flow of the Clause uh so we'll do one uh app uh that is for everyone but we can give you exercises to add more functionality to it and uh you're more than welcome to but you know uh there are we give you sample code but when you say sample answers I am not sure what you mean because this is uh apps we will give you the code yes we'll give you the code samples uh and we can point you to uh what you need to do um but uh would we be providing sample answers to all of them at the end of the boot camp uh yes I mean we have a lot of code samples I mean if you did not already realize we have more material that many of you can even process uh in a limited amount of time and then after that uh thanks to open source uh we will get you ready for open source let me put it this way this is my assessment after the boot camp you should be able to clone any G repository uh that you know that involves any kind of llm application building and you should be able to actually just get code sample not from just from data science Tojo anything that is open source anything that is out there you should be able to uh but I mean um so you mean um um um providing the full code for all the four apps I mean I I don't know what you mean by four for full code because full code could mean for some people uh the most beautiful JavaScript and the backend uh database and then everything yes I mean we give you the code samples uh for everything but how good uh you know when you say full uh it's subjective right so I want to be actually very diligent here I don't want to actually um I want to set the right expectations right so yes I mean if you're looking for a very basic app I mean you should be able to actually get the full samples course okay and uh um Adnan from LinkedIn um can we get these slides of course uh just there's a form I believe you can request the slides you should be able to get the but you have to request them uh um I think we posted um so I think the core staff can post the link out to you can request the slides um let me see uh okay that sounds good um if there are no other questions uh I am going to end this call thank you so much everyone have a great rest of the day and I am looking forward to seeing some of you at the boot camp and uh um so the one last question that came up what is the max number of people that you can accept in the online boot camp uh last time we had about 10 10 11 people and that's typically the number we don't have a lot of people online so so we are not going to have 100 people so let me how to register for the register for the boot camp uh I think we can you can click on the link uh here or and know just connect with us and we'll guide you uh how to how to register for the boot camp if you want to be invoiced uh we are vendors for many companies maybe if you work for a company where you know we can actually invoice the company directly as well just let us know sounds good thank you so much everyone um it was great having you all and I hope I will be seeing some of you at the boot camp