Build Your own Perplexity Lite using LangGraph, GPT4, and Tavily AI

hello everyone welcome to AI anytime channel in this video we are going to develop an application which will be very similar to perplexity so we are going to develop perplexity light so if you are not aware what perplexity is it's basically a startup it's an organization who helps you you know discover information faster okay so it acts like a search system where they have where they're competing with the likes of Google search okay or the Google products now perplexity also have features like co-pilots where you'll have you know you'll be using perplexity co-pilot which gives you like more humanik responses it interacts with you and and then it generate responses or it helps you retrieve information or search information or discover information now perplexity is very famous they also have something something called perplexity lab where you can you know Tinker with different llms both open source and close Source llms you can use on in free as well you know U at least on certain limitations are there but you can use in free as well I'll give the link of perplexity lab in this description now uh and how are we going to build is the most interesting part of the video because Lang chain has come up with the new updates they have come up with the new version called 0.1.0 so they have updated the Lang chain library and they have come up with a new vers version and you would have seen like Lang chain Community Lang chain core and all of those things right so they have made lot of changes and it will happen you know they will keep on adding new uh new uh stuffs into the library they will they will have more Integrations and more versions will release uh in the near future now in this update in this new uh version 0.1 they have come up with something called Lang graph which is really fascinating Lang graph helps you build language agents as graphs now just think of it once we build AI agent or J agent or llm agent whatever you name it the agents we have been building it for the last few months if you talk about Lang chain we used to build this in a loop so the the way we used to execute the agents were in the loop that's how we used to execute and it was we're using agent executor and all of those uh things which were related to agent executor but anywhere through agent executor it was running inside a loop which which might which will which is a way to execute agents but might not be the best way to work with the AI agents or geni agents whatever you call it now with langra you know it helps you with custom cyclical Behavior I will say there's a terminology I will explain that once we code it and we're going to code first in collab and then we're going to build it a fast API based uh application of course course that's what we're going to do now and I will show you a demo uh in a few minutes that's what are the application that we are going to build but Lang graph I'm going to cover in detail that how Lang graph is going to help you you know you can do explicit explicit planning you can also you know hardcore tools within uh graphs and you can because it's nodes based you can control it in a better way and how is it different from agent executor the way we have been using agents now so one is the langra and then we'll use gp4 as a model and then we going to use Tavi so tavil is Tavi is something where we'll use the search tool through Tavi so we'll combine Lang graph by the Lang chin new update and then we're going to combine with Tavi and then we'll use gp4 Vision as a gp4 preview model as a large language model and we're going to build this entirely uh within a fast API microservice and of course with bootstrap or whatever we'll see so let's jump in and see what are the things that we are going to build here here so if you look at on my screen I have to wear my glass by the way I can't see without the specs now if I look at here on uh so let ask let's ask our question here so I'm going to write uh okay here tell me five startups working in the field of drug Discovery okay now I'm asking this question and I'm you know expecting that I will get some answers okay uh here and I have asked a similar questions but this is for Quantum Computing so let's uh let's go back and ask the questions here also uh that tell me five startups working in the field of drug Discovery or something like that okay and then we can match the uh field now we have got our answer but let's ask perplexity as well and I'm going to ask perplexity so perplexity being a product they they have lot of addons features like you can search images videos they also have option to now generate images based on your queries and content and it gives you the answer here so if you look at the answer by perplexity it says molecule one in Polland cell voyant plim in citro and recursion Pharmaceuticals so it has given you USA Switzerland UK and Poland based startups name now if you look at our H answer it also gives you five okay uh it gives you in silico medicine that we know known for their synthetic molecule generator let me make it a bigger so you can know see it so let me just make it a yeah now if you read it says here are five startups working in the field of drug Discovery particularly advancing the field through the use of AI a sismic therapeutic focuses on Immunology drug Discovery pathos utilizes AI for precision Medicine pre- Science specializes in Aid driven approaches for drug Discovery in Silicon medicine not listed in and fifth was not list listed so let's do one thing we have a source here so let's copy the source and you can just come here and paste the source to validate it of course we can make it a hyperlink to just get the uh clickable link there now you can see it's get the drug Discovery over here maybe we have to uh look at the max new tokens also to uh get the answer in more detailed manner now we have our uh Source here and from this Source the model and the tool has able to retrieve so the agent is working now what I want to show you more interesting is we have something called raw Trace now within raw Trace you will be able to find out that what were the complete Trace okay below is the pass output we have passed the output once agent does all the executions we have just passed the result and the URL but this is how all the logs that you can see the raw trace of it tell me five startups working in the field of drug Discovery it gives you the intermediate steps I will tell you once we write the code what does we mean by intermediate steps okay uh uh what does it mean right intermediate step that is interesting because we have to append that once we are uh executing in working the graph uh then your query goes start off in the field of drug Discovery 2023 this is how the tool uses as a query this is a query it creates a Json of all the search results then we have a AI message you can see function call we are using open AI functions call over here and then it gives you a 10 top AI startups advancing blah blah blah uh and we can find out all the information here we have to we can get 10 also we can we have to just control the outputs okay uh the max new tokens this is a raw trace of it and this is the output that we have result and the URL so this is very similar to what you see here in perplexity of course apart from their other features that they provide because they are a product they are a startup that they have to do it now you are not building a startup over here right so a similar application a similar looking application and I will show you that how you can build this application that will help you discover information okay uh so with Lang graph because Lang graph is a topic that I also want to focus in this video so without any further delay now let's build this application guys all right so to build this application we going to first write the python code I will explain also a few things uh we're going to do this in Google collab and you can see I am in my collab notebook uh here and I have installed the required dependency just to save some time now Lang chain Lang graph Lang chain open AI Lang chain Hub Tavi python if you don't know what Tavi is if you come here it says say goodbye to hours of research now T helps you you know to it's like an AI M they say to get you rapid insights and comprehensive research it helps you in research on the internet if you want to search something it says tavite take carees of everything from accurate Source Gathering to organization of re research results all in one platform so it basically helps you you know we have Sur API right by Google so it's a next step of that it's basically provides a lot of other features and add-ons that helps you build Search application so it's a search API that helps you connect your llm to the web so we get the context we pass it to an llm gets a humanik synthesized response that's what the entire thing will be uh done here through an uh AI agent and of course we're going to use Lang graph that I explained so now if you come here we already have installed so we need two things we need two API Keys one from Tav so just create an account you can do it in free try it out here so click on here researcher you can see I'm on a researcher plan I can have 1,000 request so in the free plan they provide 1,000 request which is very good to build MVPs and proof of Concepts which helps you build something where you can go and S it to you in your school in your college to investors add the PC to your client as well if your data is not sensitive now let's come back here and just uh come down keep on adding sales now now the question is how you going to use the API keys so I will recommend that you use Secrets by collab you can see I have created two Secrets here these are two my environmental variables that I have stored open aore API key and tavore API key both in capital letter so caps and then I give the value over here now let's come here and just import our things that we need so the first thing I'm going to do is I'm going to get from Google collab from Google collab import user data and then we need import OS so let's get this and now I'm going to do os. inan okay so. environment you can see it's suggesting me now. envir I'm going to have open aore API key so let's get API key the good thing about collab notebook is that the generative a has been introduced within Google collab so it's recommendation is better now now os. inan again and then we do it for Tavi so let me just do tavore API key and then here it get user data. getet so that's it now we are okay to get our environmental key from the secrets that we have now let's keep on adding few more sales here and let's get Lang chain UTS done so from Lang chain dot uh okay let's first get input Hub now I'm going to also show you so there are two things if you want to use Lang Smith if you want to track this entire you know uh exercise you can use lsmith if you have exess to I am on wet list but if you have access you can see I am on Langs Smith and I'm going to use this prom template so the reason I'm using input Hub there can be two reason one is I can directly pull the prom templates from this Lang Smith uh Hub that we have and the other one is you can also uh have all the tracability you can see there's something called you know you have API key right now I don't have because I'm on weight list but once you have the API ke you can use it for the tracebility as well now I'm going to use this Harrison Chase open AI functions agent by the way if you don't know Harrison Chase he is the creator of Lang chain so credit goes to him to you know to revolutionize this community of generative AI the Fantastic guy so you can see open AI functions agent now this is what we're going to use that's why I'm using uh Hub as well so Hub is done now the next thing I'm going to do is Agent so from Lang chain do agents I'm going to do import create underscore open AI underscore functions agent so I'm going to use functions agent it's not function it should be functions yes and the next is from langin dot not langin linore open AI chat models open AI do chatore models okay correct and then I'm going to use here chat open AI so I'm going to use the chat thingy because you can also bring memory and all into this and build a conversational chat bot but that I leave up to you if you want to extend this further now from Lang chain underscore Community underscore not underscore do tools and then you import from the tools uh Community tools not is there there should be Tavi just a moment tools Tavi tabiz sech I not recommending me so that's probably I'm just thinking if that's the right module I'm looking at and then you import the Tavi uh search results but what's wrong in this Lang chain community. tools under score tabiz search Community tools and anyway no more module name Lang chain uncore community. tools. tavore search this is so strange I didn't install uh I have installed Tavi let me just do see that tabiz search Lang [Music] chain I have langin Community I have this search api. langin community. tools import taviz search retriever uh dot retriever Lang chain community community do tools okay let me just remove this and first import this okay this works but this is so bad I have the Lang chain Community installed or don't have oh let me install langin Community this is so bad uh just a moment I think I missed that pip install L chain how does do we have to update that now or I have community so Lang chain Community it says requirement already satisfied so why it's so bad from Lang chain uncore community now from Lang chore Community okay this is fine and then Dot and then I'm going to use tools but why okay tools is there and then I'm going to use Tavi because I also have installed Tavi oh but again it says there's nothing like that which is so bad okay import T search results I don't know what really happened I was making probably making some um mistakes in the inputs with the letters or characters whatever okay now once we we have that but this is so bad uh on my side but anyway you can also find it on Lang chain the documentation is there but we going to use Lang graph so we're going to use tabiz search in Lang graph that's what the main agenda of this video now let's have our agent definition so let me add a a mark down here and I'm going to write agent definition and in agent I'm going to write generating generating open a i functions and then it goes agent so let me just do that so now here I'm going to do tools equals so first I'm going to Define to excuse me first I'm going to define the tools and we're going to use search results so this would be in a in a python list kind of a structure tabiz search results and then you give Max I only want to return one I don't want to return more than one so you can Define that maximum results and you can see it says I need only one because I just want to show you that how you can build build it you can of course um customize it further tools the next goes with prompt so I'm going to use prompt and the prompt that I'm going to use is will be from open AI like excuse me not open lsmith Hub the Lang chin Hub so let's do hub. poool and hub. poool just give the this entire thingy over here and just come paste we'll just get the prompt from Hub now this gets our prompt next is llm so llm is nothing but the chat open AI so let's use chat open AI uh no you don't pass in llm that goes in alog together different so here I'm going to use GPT I'm going to use GPT 4 and 116 and then I'm just going to use the preview model of it so I'm just going to use the preview version of gp4 so let's get it here now if I want to see what is prompt you can just get the prompt it will explain everything that how the prompt has been you know used here so this is a prompt that we're going to use and it has all the descriptions the human masses thees blah blah blah everything right so that's what it is now the next thing is uh we need to create a runnable agent and where we'll use all the three variables that we have defined Above So for that let's create agent runnable and in this runnable I'm going to use create open AI functions correct this is right now you pass prompt llm tools okay so this is what you do but uh let's keep it in more uh okay let's keep it in more order llm tools and prompt so this now becomes in order LM tool PR okay uh agent runnable now if you print agent runnable so let me just print it it will show you assigned runnable and you can see the agent scratchpad chat from template so basically it's using something called LC Lang chain explation language I have a video on that a very explainable video on LC please watch that video as well because in the next version if I was reading at the langin recent blogs and they are already working on the next version where they are going to rely more on LC based you know uh syntaxes of you know creating this library of the beautiful library of Lang chain okay uh in the upcoming versions so should learn how to now work with LC rather than the Legacy Way of working with Lang chain now agent runable is fine so the next is we need to pass through right we need to pass through this so this is rable so for that let's import from Lang chain underscore core we're going to eat it from core so now they have three things lch core Lin community and you can find it out from open a lch as majorly we talk about open AI there are three things that you might use in your application one is langin community where you use all the open source based tools you know which has been created by Community like Pi PDF and all of those things now you have langen core which is the core thing of langen which is more Associated towards Lang chain library or the orchestration Frameworks like agents and whatnot and chains and stuff the third one is open AI as a separate library that they have created because people rely more on open AI to build gen applications when I'm talking about closed Source now Lang chain code. runnables and this is also an important thing guys to learn right don't unnecessarily Force open source open source open source right because when you go to Enterprises because they're already working with a lot of hyperscalers if you work for bigger clients they're working on with a lot of hyperscalers like Microsoft AWS gcp and they have all the credits and all to experiment and do those things also in a very secure environment right to and if even if you look at the ELO rating right now uh no no open source model has surpassed closed Source models and even the Mistral proprietary which is a proprietary model by the way by Mistral which is a smaller model in 7B category which has surpassed a few of the closed ones what I'm trying to tell you that don't unnecessarily Force things to your end user or client hear from them what they also suggest if they have the knowledge but if they go up on to you and if data privacy or protection is a problem then you look at the uh open source stuff and also with uh if you want to uh do a bit of unit economics or fops part of it then also do a comparison between the open source and the closed Source stuffs but anyway that's for a different video now runable and then you go pass through over here so reenable pass through okay so I'm just doing the Imports right now okay from L chain dot not Lang chain do core and then agents so I'm going to rely on agents import agent finish so we will use this in a conditional so agent finish okay these are the two things that I let's create an agent so agent equals and I'm going to use runnable pass through so runnable pass through and then you assign that so renable pass through and then assign and then here you write agent outcome and then this become agent runnable excuse me now let's get this and once you print the agent you will see uh runable assignment okay to again with an LC format now I'm just showing you that you should look at all these details it's it shows you everything that what in the model that has been used and all of those stuff open AI function spor this model key will also be printed and all of those stuff what prom template that have been used and all those stuff you'll be able to track it so these are the you also get this in TR tracing if you want to trace this further now let's create a service function so I'm going to do here is to text service function and let's create that now I want to create two functions so first is Define execute tools and I'm going to pass my data so the data will be from the internet and then agent action should not be caps agent action equals data. poop so I'm going to do data. pop and pass agent outcome agent action equal drop agent outcome tools to use and the tools to use should be T do name and for T in tools this is not right uh T do name then let get T4 tools for t t. name T4 will not be t for Tool this T4 T in tools this is how it should Define t in tools this is correct now here agent action and then do tool agent action. tool now this is how tools to that's the tool that we're going to use now let's have an observation observation will help you mainly when you want to do a TR tracing thing you can also use weights and bies uh when B but you can also use lsmith here now observations tools to use do execute not execute invoke so tools to use invoke and then you pass your agent action and the input of that so agent uncore action do toolor input this is correct and then you don't return because we need an intermediate step okay this is not wrong this is wrong tool input okay fine now data and then you give an inter intermediate I at steps and in this intermediate steps you provide basically not provide append append is correct but not with observation so there will be two things so you have to again give parenthesis agent underscore action and then H excuse me and then goes the observation and then return data not observation that is right so execute tools uh is good let's run this now we need one more function that's called should continue so we need a conditional here so should continue should continue again it takes data this is fine but let's write that quickly should continue data and then he will have a conditional if is instance if each instance data uh agent outcome no I have underscore so agent underscore outcome and then agent finish then you don't return this this is uh let's return exit and then have an else singy then just return continue continue okay so this is how it is let's run this now these functions are done now we're going to work with graph now so let's use Lang graph so for that this is how you can import Lang graph thingy so from Lang graph dot graph and we're going to have two things first import end and the graph thingy so graph this is what I'm going to import from this now first thing that you have to do is to create a initiate the graph so let's have a workflow and then we'll keep on adding nodes and ages so that's how you do right you have nodes and you then have ages and then you connect those ages if it's and there different types of graph unidirectional bidirectional blah blah blah right so let's let's see how we can do it so workflow and then I'm going to initiate the graph over here so let's do graph initiate that and let's add noes so I'm going to add two nodes workflow. addcore node and in this I'm going to do agent so let's do agent and then my agent is nothing but the agent and the other one is uh tools so these tools are nothing but the execute tools now these are my workflow that I have now for when you work with graph you need an entry point right in that particular graph and that's how the agent will also use that as an entry point to do execute your task now for this case you do a set an entry point so set underscore it should suggest me it should not s okay it's not suggesting me workflow. set entry point and here my entry point is Agent so let's get agent agent now this is my entry point now we need to add ages but that ages should be conditional ages and for that Lang graph has has a function that's called add conditional edes or something so let's let's try it out now for that I'm going to do workflow you can see it suggest me but I'm going to use conditional so add conditional conditional ages now in conditional ages this is very interesting now in conditional agent I'm going to first start with agent because that my start node so let me write it over here start node this is what I'm going to have now after okay excuse me after here there will be comma now then I will going to use this is continue a should continue function excuse me now we have a function called should continue right now in should continue after that we're going to have two things now first is this is right continue tools exit end correct now this is how we Define a conditional ages so let's just run this and this is done now let's see if you print workflow it will show a graph object okay now so this is a conditional age and now we'll add one age for our tools and agent so let's do workflow. addore age and then in age here I'm going to add add my tools and then I'm going to add my agent now this is done now let's go into the execution part of it so for that I'm going to add a few thing here execution blah blah blah and then I'm going to use the invoke function so chain. invoke so we have a chain and now let's use the invoke and the invoke this is how it goes you take an input and then you write your uh for example tell me five startups working in the field of sustainability I don't know if the spelling is right but whatever okay now after that you need an intermediate steps that we going to append that right now intermediate underscore but this goes inside uh two intermediate steps and then you need an empty list that will get appended that's what it is let's run this name chain is not defined it's a chain okay because I forgot to compile the graph so sorry now excuse me now we have to that's why we thinking where is the chain now we have to comp compile that right so workflow it has a compile function so let's use compile and then just do that oh this is so bad on my behalf it says agent outcome where the hell I have used agent [Music] outcome ah this should be agent outcome my bad again you have to build the graph by the way so you have to run all the sales and again you have to compile that if you directly make because functions are being used so let's do that and once you run it it will take a bit of time because it has to go on the internet get the right set of information with that will be a garbage kind of information because you canot directly use that you can directly return to the end user then you that will be stored uh you know context the context will be fed to a large language model large language model will synthesize the response and then return a humanlike response and we got our output over here so if you look at output this is the output that you have to focus so let me just copy it and show it to show it to you okay maybe I'll open Word file to show that to you so let me open a word file and show it to you that how you know we can do it now uh let me make it a bit bigger so control+ okay I was not aware that control+ doesn't work here but anyway uh what I will do I'll make this size bigger so you can read it okay now you can see it says here are five sustainability startups that you may find interesting and you get this uh please note the specific names of the startups are not provided directly from the search result but you can find a detail list of the top sear name startups sustainable review article explore the top sustainable startups blah blah blah spotlighting 33 companies that are craving a Greener and more sustainable path for the future now the startups are likely to offer innovative solutions and address blah blah blah so you get some output and also gives you the link so this is the link that it gives you now you open the link here it will take you to this wage page not found which is surprising to me okay but probably now this is the startup my bad okay so I think that I was in logs sorry so it also gets a log but you don't have to look at the log you have to look at the URL and then you have to look at the content but this is the output that you have to look at it didn't generate the name of name of it but it gives you the answer and the link to look at it but we'll see we'll ask more questions so let let's ask some questions like uh how to become an investment advisor in 6 months in 6 months I'm asking this this is what I'm asking now how to become an investment advisor in 6 months and let's see what we get over here now if we are using Tavi here so Tavi provides you 1,000 request in free you can hit the API endpoint 1,000 times okay now that is very good to build you know proof of Concepts and MVP now let's wait for this and what we get maybe we have to increase the max new tokens to you know maybe you know a bit big uh be the bigger context also to get fit in what we are getting from the internet but you can do that you can customize it let's wait for the response and then we see what we get here let me bring up my [Music] monitor and now we have got something here what what did we get uh to now you get the output it says becoming an investment advisor typically involves a combination of Education passing regulatory exams gaining experience so on and so forth the specific requirements blah blah blah it gives you a very good poins over here right in the first month second month third months regulatory exams in four months five you have to look at the licensing in month five and six you have to look at registration with the firm or something like that it gives you the entire you know uh road map that how you can become an investment advisor right in six month now let's take this code and build it in uh very quickly in a fast API app I'll show you I already have the index file ready but I'll show you uh quickly I'll explain that in the next few minutes now let me just open in vs code I go back here and then I need cd. V EnV and scripts and then do/ weight couple OB City dot dot and clean this now let's close this over here now if I look at how to we're going to use fast API as a back end you can use flask as well you can use node as well depending on that and you can for the front end we are using bootstrap 5.3 you can use react as well but I am not a react developer so I'm not doing that now if you look at here what we are doing I'll first show you up front end quickly we have using the CDM from bootstrap 5. three this will be available to you so do not worry then we have a CSS styling done to get a bit of perplexity like uh interface at least for the first one then we have you know an input box where you get your ask anything like you know you have an input ID search box basically this Dev element where search and then you you know you have loaders Etc know that loading generating whatever the answers part where we have the raw trace and accordion which is a collapsable accordion over here we have changed the color of it so some CSS styling then we have the answer card which shows the below past output which is result and the URL now we using the script TX you know the script CDM for bootstrap sweet alert Etc font awesome we have that font for collapsable headers and the icons now this is a JavaScript function which is a vanilla JavaScript where we you know using document. getet element by ID of the query so query is nothing but the query that we write in app. PI from the form of fast API then we are using a Fitch over here and you can see method of post you know getting the form data this is going to be our API endpoint name process query so this is what we're going to do and then you just passing the output and sewing it over here simple so I hope you understood the HTML part of it I don't want to explain that in detail now you have to create a templates folder keep your HTML into that and then you static file if you want to keep logos and stuff let's create uh app.py file I will show you the requirements these are the requirement that you need to work with this application fast API uon python. EnV and ging to are the added ones we already have this from Google collab now in app.py what we're going to do is we're going to get few things so let's first get uh let me go back and just show you you can just show it to from here let me just go I'll just paste I don't want to write every line of code because you already know now that I'll just copy and show you that how you can uh do that I'll explain the code or the fast API now let's go on top if you look at here these are the things you already know you have seen that on Notebook fast API we are importing and we need a few things from Fast API we need request and form form is to get the query from the end user and then we have Jon response we are using to get justable encoder to pass the out put for Jon realization and all of those stuff and then you have redirect response that redirect to Local Host 8,000 SL dos to a Swagger UI or it redirects to you to you to the index.html now Evon stuffs are happening because that's a server that helps you run fast API applications Json and Bas 64s are happening you get the import thingy done you get the this code is exactly the same I'm not I haven't made any changes till here to compile I'm not using invoke now what I did I have written a mi Ser serializable function because to pass the raw Trace Json was not calized that's what the error I was getting so I made a function that helps you handle the raw traces the in this we get from internet or the output that we get now there's a function that we'll use to pass the output basically now this is a if you want to run this this is the same thing then you initiate the fast API app like this app equals to fast API then you mount your static directory you have your logos and stuff that's why you mount template where you keep your HTML files canot template I have a get endpoint okay that basically takes you to index.html when you do Local Host 8,000 that's what it is and then this is your main function a single function process query that you saw in index.html JavaScript function as well uh in in Fitch now here we have query string which is a string query and the form then we have our intermediate steps if you remember there and then we are using chain. invoke after that it's all passing the way you pass your data to show on the front end screen that's it now this is how you do it and you can just run it over here and I can just just show you quickly how to run that because you already have seen but I'll just so now once you do Port because I'm running on 80001 because I already have a Docker container running on uh some think I'm running inside a Docker container on that Port 8,000 I cannot utilize that uh let me open uh Local Host 80001 your knowledge begins now this is what we wanted to build guys right we started our uh experiment in collab and now we built here we already know the output that you saw yeah but this is what it is now if you go and ask the similar kind of questions let me ask uh tell me uh five startups working in the field of artificial artificial intelligence sometimes it might give you Json error so you have to handle that I I was still trying to figure it out so if you get some Json error that which is not realizable then you have to handle that and this is something that I leave up to you if you get those error if you get that error please provide that in comment box I'll try to help you but sometimes you might get it now you can see here is the startup that you get uh AI squared AR AI AR AI is a fantastic startup you know they have come up with that uh ml observability okay you can know a bit of explainability and observability part of llms that they do then we have open AI the best startup that we have in right now in the world valued more than know how billion I don't forgot the exact number Landing Ai and all those things it gets it from here top 10 hottest hottest AI startups okay on the crn.com news Cloud yeah so this is what it is guys I hope you now uh and I'm have to remove this because some of you said that ring light and SP doesn't go along so I'm just removing that now I hope you understood that how you can use Lang graph that helps you build language agents as graphs not as a you know running in a loop kind of a with agent executor that's the Lang graph we use a tool that's called Tav AI maybe if you like that tool because it gives you 1,000 request which is very uh good to build you know proof of Concepts and then we use gp4 so we combine all of those things and build this application called perplexity light kind of an application which is a 10% of what perplexity does right now the code will be available and if you have any question thoughts feedback please let me know in the comment box and uh if you like the content please hit the like icon if you haven't subscribed the channel yet please do subscribe the channel share the video and Channel with your friends and to peer thank you so much for watching see you in the next one

Channel: AI Anytime

Views: 5,303

Keywords: ai anytime, AI Anytime, generative ai, gen ai, LLM, RAG, AI chatbot, chatbots, python, openai, tech, coding, machine learning, ML, NLP, deep learning, computer vision, chatgpt, gemini, google, meta ai, langchain, llama index, vector database, LangGraph, langgraph, Langchain, langchain v0.1, perplexity, perplexity ai, GPT4V, GPT-4, GPT4, tavily ai, Tavily AI, Tavily, tavily, search chatbot, search system, search app, RAG search, perplexity vs google, langchain updates, langchain new version

Id: O0fpDUwxUEg

Length: 43min 14sec (2594 seconds)

Published: Thu Jan 11 2024