​
Handling complex databases with many tables

With our current architectures, Brewit can support databases with hundreds of tables without LLM exceeding the token limit. We use RAG + LLM framework to find relevant tables metadata based on users’ questions.

​
FAQ

​
Filter string columns

Filtering stirng columns can be challenging with LLM without seeing the actual data format. To handle this, we have a feature called Distinct Values.

It essentially run the following SQL and store results in the vector store for retrieval:

SELECT DISTINCT <column_name> FROM <table_name> LIMIT 50

​
Example

If a user asks what is the average salary of software engineers in the U.S.?, the AI model may generate the following SQL

AI-generated SQL without distinct values:

SELECT * FROM salaries
WHERE country = 'U.S.'

The AI model may not be aware that ‘U.S.’ and ‘United States’ are the same country. As a result, the query may not return the correct results.

AI-generated SQL with distinct values:

SELECT * FROM salaries
WHERE country = 'United States'

By using distinct values, the AI model improves the accuracy of the where clause.

​
Suggested roll-out strategy

  1. Initial rollout: Start with a small subset of data power users and a subset of the database to test the system and gather feedback.
  2. Gradual expansion: Gradually increase the number of users and data scope to ensure the system can handle the load.
  3. Full rollout: Once the system has been thoroughly tested and optimized, roll it out to all users.