Still Looking at Text-to-SQL? You're OUT! It's Now RAG2SQL! (Step-by-step)

In this episode, join Angelina and Mehdi, for a discussion about RAG2SQL or RAG2Database, which is an extension of the hot Text-to-SQL concept last year. A lot of the Text-to-SQL apps failed, why?
Check our episode out!
Who's Angelina: / meetangelina
Who's Mehdi: / mehdiallahyari
00:00 Intro
00:07 Why is talking to your database business critical?
00:45 What is RAG (retrieval augmented generation) to SQL?
00:58 Introduce Vanna
01:21 What kind of database can it talk to?
01:49 Text-to-SQL?
02:10 How's RAG-to-SQL different?
04:00 What does the RAT-to-SQL training phase do?
05:52 Is there real training done?
08:44 Step-by-step tutorial (Jupyter Notebook) of how it works with a SQLite databae
09:55 How to download a SQLite database on Kaggle
15:10 What more does it offer? (👉 shh... free UI! and more!)
15:47 Compare with DataDM
16:26 Performance
19:31 If you are a company with bunch of predefined SQL queries, what do you do?
20:10 Are challenges like loss-in-the-middle still applicable?
🦄 Any specific contents you wish to learn from us? Sign up here: noteforms.com/forms/twosetai-...
🖼️ Blogpost for today:
DataDM: open.substack.com/pub/mlnotes...
AI SQL Accuracy: Testing different LLMs + context strategies to maximize SQL generation accuracy: vanna.ai/blog/ai-sql-accuracy...
🔨 Implementation: Vanna AI: github.com/vanna-ai/vanna
📬 Don't miss out on the latest updates - Subscribe to our newsletter: mlnotes.substack.com/
📚 If you'd like to learn more about RAG systems, check out our book on the RAG system: angelinamagr.gumroad.com/
🕴️ Our consulting firm: We help companies that don't want to miss the boat of the current wave of AI advancement by integrating these solutions into their business operations and products. www.transformaistudio.com/
Stay tuned for more content! 🎥 Thanks you for watching! 🙌

[@briancase9527]

Thanks for this introduction. This is evolving rapidly, which is great, but it's hard to keep up. :)

[@adnanrizve5551]

You two are doing great job! Keep sharing your knowledge

[@sneakeraplus]

Thanks for this. How about more complex/indirect questions that require more reasoning over the schema beyond standard defined relationships for example: 1. Tell me something interesting about my process/data 2. Which of my employees have shown the best the best collaboration practices and what are they doing right?

[@yuzual9506]

salut ! Thx for all your vids, i'm french and i want to use rag in research, i learn a lot with you.

[@kenchang3456]

Indeed, thank you for this introduction. Looks like you don't need a GPU to do the embeddings when "training". Very timely for my POC, much appreciated.

[@johnkintree763]

Excellent. An extension of this would be RAG to SPARQL.

[@joneskiller8]

That would be amazing.

[@kenchang3456]

Just curious, at 15:15, for the user question "what are the top 3 cities with the most customers" did Vanna formulate the SELECT based the DDL & DML that is in the vector store on its own or was there a view that already existed that it used. If I had just tables and I loaded that schema in the vector store would Vanna be able to figure out the joins? That would be impressive. Of course, I could experiment myself but thought I would ask.

[@priyatosh-ig5eo]

I am working on the SQL chatbot and have used Vanna AI and trained my own model, sorry to say but response are not that good and as expected.

[@AliAlias]

Great job 🎉🎉🎉 Can integrate with local open source LLM using llama?

[@atrocitus777]

have you gotten an answer for this? I work with proprietary data so just making a call to an endpoint of any kind on commercial internet is a NO for me.

[@AliAlias]

@@atrocitus777 not yet

[@__________________________6910]

What a thumbnail

[@gazart8557]

Does this mean data analysts will be out of job?

[@TwoSetAI]

Probably not in the near term, but it should be able to empower DS/DAs' work. If you are in DA/DS roles, we recommend trying these tools out!

[@luka1790]

I gave it a shot, but unfortunately, it didn't perform well. I wanted to know the number of orders in the database, but it tried to query SELECT COUNT(*) FROM orders instead of SELECT COUNT(*) FROM public.order. It seems it's not fully reliable yet.