Ask HN: How to augment LLM/GPT with structured data in a safe way?
I have seen many applications augmenting LLM/GPT responses with unstructured text data using frameworks like LangChain/Vector databases. The idea is basically querying similar document chunks and plug them into the prompts.
But for structured data stored in SQL-like databases, there doesn't seem to be a widly adopted way.
I have seen exapmles from the LangChain library using GPT to generate a SQL query and execute it using Python, but this would create security issues if the LLM generates queries that tamper with the database.
Thanks for any inspirations or input.
No comments yet.