Chat with MySQL Database using GPT-4 and Mistral AI | Python GUI App

[@aristoz1986]

Great!! I will try it out this week!! Keep on going with the good stuff🎉

[@alejandro_ao]

Thanks!! Let me know how it goes!

[@BrandonFoltz]

Oh yea! More to learn this weekend! Cool to see all of these technologies merging and synergizing. Keep up the great work!

[@alejandro_ao]

Thank you Brandon! Always a pleasure to see you here!

[@avijitsamantaray8420]

Great work ❤

[@jim02377]

I just made it through the entire tutorial. As usual you attention to the details is awesome. Now I want to try to agent version and see the difference!

[@alejandro_ao]

thank you Jim, it's great to see you around here! i'll be putting a version with agents soon!

[@lookingaround1586]

Thanks @alejandro_ao. Could you make a video on implementing graphs/charts alongside the NL response? Congrats on your diploma!

[@oscarzamorapicazo9287]

Thanks Alejandro! GREAT STUFF AND EFFORT

[@AasherKamal]

The way you explain the details is impressive.

[@trideepsaha2594]

This is what we have been waiting for a time, No words only 🎈🤩..We must all do our homework. Special thanks for groqcloud.Congratulation on your Diploma AO.

[@alejandro_ao]

thank you!! very glad to hear this was useful! let me know what else you would like to see!

[@trideepsaha2594]

@@alejandro_aoFrom this side anything u want to teach. #AI #RAG #LangChain #FineTune

[@RahulM-lm6qg]

please make video for this usecase only using open source LLM models

[@homerobaroni1655]

A version using local Ollama would be nice!

[@shashankkumardubey6260]

Great project. As a beginner this project can help me to learn better. Do make some more unique projects . Keep doing great work.

[@alejandro_ao]

thank you mate, i will!

[@scollin10]

Love your walkthroughs! I’m using your method and approach to chat with a MySQL database of resumes. The eventual goal would be to build up the database of both resumes and job descriptions and have my team of recruiters be able to prompt it to optimize efficiency.

[@alejandro_ao]

hey there, that's an awesome idea! just remember that, for a real application, you should not use the `root` user of your mysql database. create a new user that only has READ privileges. you wouldn't want your LLM to accidentally write anything into your database or delete some data!

[@scollin10]

@@alejandro_ao I use a .env for now just as good practice but it is all local as I’m encountering some issues. I would love to pick your brain and chat over coffee or a consultation session if that’s cool.

[@scollin10]

@@alejandro_ao Thanks for the tip on using root! I would like to pick your brain and seek your help on my app. I'll shoot you an email.

[@Suryav12]

Great idea

[@Suryav12]

Have you done it? And what's the approach you went with. Is it something like u let the say hr upload the resumes and extract necessary information and store it in database. And then do the stuffs in this video. Is it like that?

[@vinayakmane7569]

I am loving your work bro . dont stop . keep making such unique projects

[@alejandro_ao]

thank you! there is much more coming up :)

[@OwaisBinMushtaq]

Great .... Will try to implement this week 🎉🎉🎉🎉

[@alejandro_ao]

let me know how that goes

[@mlg4035]

Congratulations on your diploma!! Great video!

[@alejandro_ao]

thank you!

[@Sanitiser254]

Love your work man

[@alejandro_ao]

i appreciate it :)

[@peralser]

Great Work!! Thanks for sharing your time and knowledge with us.

[@alejandro_ao]

i appreciate it man. it's my pleasure :)

[@babas5990]

Congrats on earning your diploma.!!! Thank you for your excellent video tutorials.

[@alejandro_ao]

thank you!!

[@muhammadqasim6524]

Congratulations on your Diploma. 🎊 Enjoying your videos.

[@NasserTabook]

Great tutorial, Thank you for the hard work

[@alejandro_ao]

thank you, i appreciate it!!

[@matiasparouy]

Thanks Alejandro for sharing! excellent video! 👏

[@alejandro_ao]

thank you man :)

[@MarioLopez-bm9mf]

Great, thanks for sharing. Quick question: Can it be used with large databases that have more than 100 tables? What do you recommend for handling large databases?

[@adityabhagwat6965]

Great content! Appreciate you sharing. Excited to give it a go!

[@Jay_jadan]

Congrats on the diploma sir 🎓

[@alejandro_ao]

thank you sir 🫡

[@songfromktown]

Thanks for this great tutorial. In case I would like to have some plot (bar or line charts of any column of a table), how should I approach ?

[@DoggoDebugger]

Great stuff!

[@alejandro_ao]

Glad you enjoyed it!

[@ethanxsun]

So cooooool! You are hero. Talented

[@alejandro_ao]

glad this was useful! keep it up!!

[@SibiP-b9x]

is it possible to add a visualizations ?

[@ArusaKhalfay]

I just built this, it does basic stuff and is great but if you notice and test it out with the last user question it actually gives the wrong response and does not correctly map the artist ids. Do you have ways to improve this performance. It makes mistakes with writing simple joins too and throws an error?

[@luismario6808]

amazing video, thanks!

[@alejandro_ao]

Glad you liked it!

[@claudiogandolfo2627]

Hi Alejandro! Thanks for the video and congrats for your achivement. Here is my little contribution to your great viedo. It seems that the example SQL query for question: "which 3 artists have the most tracks?" is not right (at minute 28:27). One solution could be "SELECT Header.Name, COUNT(*) as track_count FROM (SELECT artist.Name, album.ArtistId, album.AlbumId FROM album JOIN artist ON album.ArtistId = artist.ArtistId) AS Header JOIN track ON Header.AlbumId = track.AlbumId GROUP BY Header.ArtistId ORDER BY track_count DESC LIMIT 3;" 😉

[@openyard]

Thank you. Another video with great content.

[@alejandro_ao]

thank you!

[@ilsms]

Thanks for sharing ! The best video

[@alejandro_ao]

it's my pleasure!

[@victorchrist9899]

Congratulations on your diploma

[@alejandro_ao]

thanks!!

[@AmirShafiq]

Good stuff and great explanation

[@alejandro_ao]

thank you sir 🫡

Thx Alejandro , great stuff like always .. crystal clear 😎. I changed your code a bit to use it with a postgre database, works like a charm !✊❤‍🔥

[@alejandro_ao]

that's so cool! which adapter did you use to connect? psycopg2?

@@alejandro_ao yes this one

I started to push the prompts a little further (formatting, tone, etc...) it's a really good base to work with. Thanks again for this starter kit 🙏💪

@@alejandro_ao did you ever try with a mongodb?

[@alejandro_ao]

@ so glad to hear this!

[@tinsp253]

What a great work. Thank you. If the database tables don't have proper names, I guess the system won't work am I right? When I looked into my client's database, they used silly names that didn't represent the content of the table. How can I resolve the problem? Thanks again.

[@andersonkoh1382]

Do you have a series for beginner to learn how to build custom theme and create pages with these? I'm used to build with the builder in WordPress.

[@TYTennis]

Hi, this is great thank you so much I'm learning a lot from this. I see you said this wasn't production ready and that improvements could be made. What improvements do you think you would make and why would that be an improvement? Thanks again for this video, you're very ahead with tech :)

[@ishanagrawal396]

THanks! How can we us multiple database?

[@delgrave4786]

Hey alejandro amazing project! I implemented this made a bunch if changes to it too. I had a request, do you have a video about creating and working with agents using open source llms like groq. If not can you make a tutorial for that? Most videos i see always use openai and i just am not able to implement them with open source llms. And openai refuses to take my debit card

[@LukeS-e3m]

Are you able to create a video where you're searching through multiple tables? Or would this approach work for this too?

[@RakshithML-vo1tr]

Any plans of making videos with the help of llama 405B parameter model

[@AJITKUMAR-k9k2g]

Are you using any extension which gives you autocomplete code suggestion?

[@kingfunny4821]

Thank you for your wonderful efforts and excellent explanation Please, I would like to ask you whether it has been explained how to make a conversation with files without using the Internet, that is, it is only local

[@ritishajaiswal9918]

Hey Alejandro, Thanks a bunch for the helpful video. Could you make another one where you show us how to use a model we've downloaded locally (Can be a quantized one)?. Also, it'd be awesome if you could include Vertica database as the DB in the process. You don't have to go through the whole process again, just maybe explain how to connect to the database using the local model. I've heard there are models like SQL-coder-7b that are designed for translating plain English into queries. It'd be really helpful if you could check out a few of these models and share your thoughts on which one would work best in this kind of situation. My organization is pretty cautious about sending sensitive data to LLM model APIs, so it's important for us to be able to do this kind of thing locally without relying on APIs. Looking forward to the next video. Thanks again for all your helpful content!

[@laurarusu1989]

Hi, I did this with llama3-7b. Actually the code is the same one from the video. First, you install the local model. in my case I installed llama3-7b by running in Terminal 'ollama pull codellama:7b' then I run 'Ollama run code llama:7b' (you can go to the llama site and see the actual command) then when I set the llm I used 'from langchain_community.chat_models import ChatOllama llm = ChatOllama(model="codellama:7b")' . everything else should be the same. Langchain supports many different llms, you could see the documentation and check if the llm you're interested in is available and use the corresponding module. I hope this helps you :)

[@blackpanther3005]

Hello, good day, excellent video, sorry, how much VRAM did you use for the project?

[@alejandro_ao]

hey there, all the completions of this app are done in remote servers, through the providers' APIs so they are the ones dealing with all the processing power. you could run this in an old raspberry pi tbh 😎

[@TheAbanik]

Amazing video as usual, very helpful. Could you explain why you decided to use GPT 4 instead of 3.5 Turbo ? Is GPT 4 better in crafting the sql queries based on natural language ? Also can you try creating a video based on langchain agent to showcase a use case where user asks a question, the response can be either in a document/pdf or from a database and agent needs to figure out the correct way to respond.

[@alejandro_ao]

hey there, sure. i used gpt-4 because i needed its 128k token context window to be sure that it could ingest the entire database schema. but actually, when the schema could fit in GPT-3's context window, i found that it actually worked better. i will be doing videos on agents very soon!

[@niyetheg]

I need your help, I followed all the steps and I am at the last part where I ask questions and get feedbacks but I keep getting this error validation error even though I have included the API KEY. for ChatGroq __root__ Did not find groq_api_key, please add an environment variable `GROQ_API_KEY` which contains it, or pass `groq_api_key` as a named parameter. Please what can I do?

[@gianni4302]

print under it to see if its accessing the Groq qpi key, or just input the key and see if it works. if it does its an issue with the parth to your .env. make sure to 'load_dotenv()' at the start instead that may help

[@bobcravens]

I’ve been enjoying your channel. I wonder at what temperature there is a probability of the LLM dropping tables from your DB ;-)

[@alejandro_ao]

oh yeah, absolutely, please DO NOT use this with a mysql user that has write privileges

[@AyomideKazeem-g7n]

Hello, great tutorial as always. I am completely new to this, so after running the code and changing a few things I keep getting this error "AttributeError: st.session_state has no attribute "db"" and I do not know how to solve it.

[@MamirZhan-de7fv]

I have been following your tutorial, it helped me a lots! Thanks for sharing. All these LLM models are greatly depends on sementic meaning of the tables and columns. But in my case, my database came from Open Way loan management system and most of my tables and columns name doesn't have semantic meaning. For example, the transaction table called DOC instead of transactions. In side DOC table, non of the fields name are self explanatory. As a result, most of time, my model can't answer my questions. What should I do?

[@showbikshowmma3520]

Firstly, love you 3000 times! I have a question. I want to build a chatbot using an open-source language model like LLaMA 3 or another available large language model (LLM). How can I integrate the LLM with my MySQL database so it can answer questions based on the information it finds in the database? Additionally, I would like to integrate an API into the LLM because I have a hosted Python backend server. Through the API, the LLM will also be able to respond to user queries. I believe u will help me out in this case..

[@arvindprajapath2315]

How are you getting those code suggestions or auto complete feature when you are typing the code in vs code. Can someone tell me which extension is this

[@RaushanKumar-ut2ke]

Hey alejandro. Do we have any method to connect multiple schema for database chatting

[@openyard]

I followed this video and put together a RAG solution uses LLM to query a PostgresQL database (llm = ChatGroq(model="mixtral-8x7b-32768", temperature=0);) and it works well. But when I try to query a graph data implementation in PostgreSQL consisting of two tables "node" and "edge" the generated SQL queries are error prone. The query should use WITH RECURSIVE CTE queries, but somehow the LLM generated queries always have SQL DML query error and mostly do not RECURSIVE quries, Is there a way I can send examples to the LLM model through some hand crafted pairs of prompt and PG query examples?

[@aadilgani5528]

Congratulations on getting your diploma, can you please make a video On Chat with SAS (clinical data .bat files) database using open-source LLM.

[@alejandro_ao]

I have never used sas, but a senior technical architect there implemented an agent that interacts with sas data. Maybe this can be useful to you: www.linkedin.com/posts/bogdanteleuca_how-to-create-your-custom-langchain-agent-activity-7161603166952734720-jqRo/

[@LieutenantGeneral]

I tried to train a bert model on Spider dataset but realised that I need a dataset that also consists of DDL commands, any suggestions or do I need to nake a custom dataset?

[@LieutenantGeneral]

Also to mention that my choice with bert was not good, so now I am going to use T5 model.

[@SatyendraJaiswalsattu]

Nice tutorial, please make one video - how to entract with my excel sheet data

[@alejandro_ao]

great idea!

[@thiagomantuanidesouza136]

What to do if the columns contain underscore in the names? Is there any configuration to solve it?

[@thangarajerode7971]

Could you please create the video about how to deploy the steamlit app into langserve?

[@aniketdeshmukh5776]

Thanks Alejandro, for excellent video! I have one doubt, when we are using openai key with our sql database. are they able to access all our sql database data? If yes, then please let me know, how to prevent that. If No, then it's fine. Thanks for clarification in advance...!

[@alejandro_ao]

hey there, great question! according to the openai's docs and privacy policy, they don't keep logs of your prompts for more than 30 days. and they do not use these prompts to train or fine tune the model. so if that is your concern, you should be fine. however, you may absolutely do not want any of your data to go through the openai's servers regardless of their promise of privacy. if that is the case, here is some info about the implementation in this video: openai's servers only see what you add to the prompt that you sent to them. in this case, we are sending two prompts to openai during our process: 1. firstly, we ask the LLM to generate the SQL query. in this case, we are sending the schema of the database in our prompt. the schema itself does not include the data from your database, but it does say what kind of data your database contains, such as the names of the tables, the data types, etc. the LLM will use this schema to generate the SQL query. 2. secondly, we run the SQL query towards our database. this operation happens only between ourself and your database. openai is not involved here. 3. thirdly, we take the results that our database returned and we send them to openai for interpretation. here is another step where we are sending data to openai. so in short, in this tutorial, we are sending two pieces of information to openai: the schema of your database and the results from the query. if you are not comfortable with this, you can always run a local model using ollama! this will work great :)

[@aniketdeshmukh5776]

Thanks @@alejandro_ao, For excellent clarification. Really Appreciate👏👏

[@zefur321]

What is the extension name that you use to predict the next steps of coding (i see it will show as gray color) in VS Code?

[@alejandro_ao]

that’s gh copilot

[@RlUpadhyay-c6o]

will be work for large database like I have 1200 hundred tables in my schema??

[@juanjubia]

Could be possible to do something similar but with MongoDB community?

[@redo22011]

Thxs for sharing, its a great work!!! can you add similar example for amazon bedrock Claude 3

[@danilabelokon4202]

Is it possible to build a chatbot that can handle updates to queries? For example, I want to create a car parts inventory system where I can add and remove parts using voice and chat. Additionally, I want the chatbot to provide functionality for asking about available parts.

[@alejandro_ao]

hey there, it sounds to me like you want an agent that is capable of performing CRUD operations on a database. that is possible, you just have to create some tools (python functions) and bind them to your agents. you can use langgraph for that!

[@MohitThehuman]

Hi Great video Thank you I have one question how will we deal when there will be like 100 tables in the schema Like it will lead to the context window issue Do we need to add -vectordb to first search and filter out the relevant information then pass it to llm Any comments ?

[@AIWALABRO]

I love this video! can I put it into my resume? In the end, you said that in real life we do things by "agents" instead of "chains". here we did it with "chains". 1) so for the production-ready how we approach it are we used chain or agents? 2) can you make a video by using agents?

[@alejandro_ao]

hey there, absolutely! feel free to add this to your cv :) 1) i would test both. the only difference is that an agent would probably be better at processing your query. it would be able to check the result from the mysql database and reformulate its sql query if the result does not answer the user's (your) question. best would be to test both and see which one performs better! 2) coming up!

[@AIWALABRO]

@@alejandro_ao thanks for your response! eagerly waiting for the agents video in mysql, we can say mysql part-3.

[@JuanPalacios]

excelente informacion!!!!

[@alejandro_ao]

gracias!

[@devizl]

thnx pro for everything this was amazing iwanna aske u a Q if i want to use my llama 3 local model how can i integrate it in the code i use: from langchain_ollama import ChatOllama llm = ChatOllama(model="l------", temperature=0) do i need to import url and if so do i need to put it in .env file and what is the correct way and thx again

[@alejandro_ao]

hey there, did you manage to get this working? as far as I can see, all you need to do is start up your local ollama service and you're good to go for using the langchain integration, right?

[@Parthi97]

can you show me the steps when my streamlit app is hosted on cloud so that when the user enters connection details my web app should be able to connect

[@Shivang0369]

Hello brother where have you used LangChain's API? You mentioned it in .env file ? Also i am using the OpenAI model 3.5 for which i am getting below error: openai.BadRequestError: Error code: 400 - {'error': {'message': "This model's maximum context length is 16385 tokens. However, your messages resulted in 16734 tokens. Please reduce the length of the messages.", 'type': 'invalid_request_error', 'param': 'messages', 'code': 'context_length_exceeded'}} kindly tell me what should I do? I am planning to build this chatbot at enterprise level, which have larger database. Keep uploading brother, great explanation

[@fozantalat4509]

Can you create this app tutorial using agent instead of chains, that would be really awesome.

[@alejandro_ao]

i will probably do something like this using CrewAI

[@1sanak]

Grats on the diploma! Can I use the same method to connect to a MSSQL DB?

[@alejandro_ao]

thanks!! i heard some guys from the channel were trying to do this. i haven't tested myself. but look, apparently you can pass in a driver just like we did here with mysql, but for MSSQL: docs.sqlalchemy.org/en/20/dialects/mssql.html#module-sqlalchemy.dialects.mssql.pyodbc

[@djaz7347]

Hi Alejandro , is possibile to connect a DB2 database through jdbc?

[@yugeshkaran7547]

Thanks for the video bro . Now I have built and hosted the similar chatbot on streamlit cloud community, but I have encountered an error with connecting with MySQL database. I have crossed my database server but still i haven't resolved it. Could please guide me what the issue might be??

[@RaushanKumar-ut2ke]

All the methods I saw is to use single schema but I didn't find any workaround for connecting multiple schema for database chatting

[@AIgencify]

Next video with agents!!

[@alejandro_ao]

yes sir

[@jimerlozano6523]

interested in more about sql.

[@alejandro_ao]

i will work on it !

[@nguyenduythai4585]

If using open_ai_key, will the data in the database be leaked, how can it be secured?

[@Ganitham-GenAI]

Will the test results and the data be exposed to LLM or only meta data

[@alejandro_ao]

yes, all data goes through the LLM. if privacy is your concern, you can load an open source model locally, such as llama3

[@udaynj]

Does this mean that OpenAI and/or Langchain have access to your database schemas and also the data generated? That could be a huge security risk for most companies.

[@akshaysangwan2172]

The code you taught works properly for small queries but fails to write complex queries and provides incorrect information. Can you help? I'm using ChatGPT-3

[@JoEl-jx7dm]

Hey i have issue hosting on streamlit cloud, could ya try?

[@PhamDuc8504]

I want to ask how the VSC IDE can suggest the remaining parts of the code ? Thank !!!

[@alejandro_ao]

that's github copilot!

[@stanTrX]

Thanks but traditional forms to pull data from db's would be probably more efficient.

[@avinashnair5064]

Hey, actaully i want to crete a RAG applciaiton which can do all the stuffs like code agent + SQL agent + a normal document retriever how can we do that'?

[@alejandro_ao]

you would need to create an agent for that! then create the tools for code, sql and rag. or you can use a multi-agent system, like laggraph or crewai!

[@avinashnair5064]

@ i want to keep it completely open source and everything should be self hosted could you please propose a simple solution?

[@alejandro_ao]

@@avinashnair5064 both crewai and langgraph are open source. you really can’t go wrong with any of these! langgraph is great for more controlled workflows. crewai is better for more creative automations, like emulating a human team of specialists in different fields

[@saibaldasgupta1]

Is this code can be use for postgresql?

[@vincentfernandez7145]

You need to use psycopg2-binary instead of my-sql-connector

[@tofani-pintudo]

Yes, you just need to change the db config.

[@manikandanr1242]

If we ask questions from out from db example user enter hi what will be it response

[@alejandro_ao]

hopefully the overall chain would be able to get that you were actually just saying "hi". but in order for your program to decide whether or not to call the database depending on the user question, you might need to create an agent or a bit more sophisticated chain

[@王君淑]

Hello, greetings and congratulations on obtaining your diploma. Once again, I would like to ask you a question. I may need to analyze MySQL data. The first step is to execute sql_chain and run SQL statements, and the second step is to add an analysis chain. The problem I encountered is: when I add self-awareness knowledge to prompt words, and when I ask self-awareness questions, two chains need to participate at the same time. How can I control the first chain not to execute or have any other way to only run one chain, which may make the program more powerful. Let me give an example: User: hello! AI: hello, can I help you? When I asked this, he didn't need to execute SQL statements, but now he still does. So there are the issues mentioned above. I hope you can understand my language, because I am Chinese and this question was translated through translation software. I do not know English. Wishing you again.

[@alejandro_ao]

Hey there. If you want something more interactive, that is able to respond to other questions and not only those strictly related to generating a SQL query, you should probably implement an agent. LangChai makes it very straightforward to implement agents: python.langchain.com/docs/modules/agents/. You can also use CrewAI to create a team of agents: docs.crewai.com/ I hope this helps!

[@王君淑]

@@alejandro_ao Thank you very much. I have benefited a lot from watching your latest video.

[@tapanpati9452]

cool,bro i have a scenario to create a chatbot using 36,000 pdfs ,do you have any idea,but rag is not giving accurate answer.

[@stanTrX]

How those code tips are popping in vs code? Does anyone knows?

[@alejandro_ao]

hey there, that's github copilot. saves a lot of time :)

[@who-kv6fe]

After deploying how can i turn the host from localhost to a public server/web host? Any solutions?

[@alejandro_ao]

hey there, take a look at this: kzbin.info/www/bejne/bWXGZH6Xdr2DrM0si=EYXg71ydGQzZQZmc

[@tapanpati9452]

Can you please make a video how to use langsmith?

[@alejandro_ao]

definitely, i'm working on it!

[@VoltVandal]

congrats !

[@alejandro_ao]

thanks!

[@nguyenduythai4585]

i want it to print the message that the question is not related to the db. Help me ?

[@alejandro_ao]

hey there, maybe you can add to the final prompt that "if the context does not contain the answer to the question, please say that the information could not be retrieved from the database", or something like that!

[@nguyenduythai4585]

@@alejandro_ao you can help me with code...

[@nguyenduythai4585]

@@alejandro_ao can you help me with code

[@nguyenduythai4585]

@@alejandro_ao you can write code here.

[@gianni4302]

Great tutorial, but i keep running into error 400 due to token counts. Any real world schema alone will bump up to almost 10k tokens. anyone found a workaround? A tutorial on this would be lifesaving!

[@VibhuSharma]

I am running into the same issue, always getting exceeded token limit error. Were you able to figure out any workaround?

[@adelinadonisa]

At 9:50 how did that SQL database chat page opened 😢? Maybe I am stupid but I can't figure it out if you run the code or not 😂

[@alejandro_ao]

hey there, that’s a good question. what i opened there is not actually part of the app. i just opened the command line of my computer and showed that i have a mysql server running locally ;)

[@adelinadonisa]

@@alejandro_ao Thank you!

[@CodexOdyssey]

please make video ai manager API

[@LadiAsrith]

How I set up langsmith? Please help me that

[@alejandro_ao]

hey there. just create an account and add the api keys. if you’re using langchain, it automatically starts tracking. if you’re using any other framework, you’ll need to manually specify what you will track