Everything K-culture — comebacks to K-beauty, straight to your inboxGet it in your inbox

METAL MEDIA

TOPDEV99999/AI-Knowledge-Management-Platform

94Jupyter NotebookMIT-0

An AI assistant that reads company documents and can actually calculate answers from them

This open-source project is an AI assistant that not only searches company documents to answer questions, but also pulls structured data out of them and runs SQL queries to answer analytical questions, like 'which contracts expire in the next 3 months?' An LLM agent decides on its own whether to search, query a database, or use other tools, letting it handle multi-step questions too.

What it does

  1. Uses standard RAG (retrieve relevant text, then generate an answer) to handle factual questions from documents
  2. Batch-processes documents to extract structured entities into a database, then converts analytical questions into SQL queries run against that data
  3. A ReAct-style LLM agent decides whether to use search, SQL, or other tools to answer a question
  4. The repo includes a backend, a Next.js chat frontend, document-ingestion pipeline notebooks, and experiment modules
  5. Jupyter Notebook-based project with 94 GitHub stars

Why it matters

It shows a practical way to automate digging through unstructured business documents like contracts, including questions that need actual computation such as expiration dates, instead of manual review. It's also a concrete open-source example of combining retrieval-based AI with data-analysis AI inside a single agent.

Terms in this repo

  • RAG · Retrieval-Augmented Generation: search relevant documents first, then generate an answer based on them
  • ReAct · An agent approach where the AI alternates between reasoning and taking actions (using tools)
  • SQL · A query language for retrieving structured, table-like data from a database
  • entity extraction · Pulling out meaningful pieces of information, like parties, dates, or amounts, from documents
  • vector-based retrieval · Search that turns text into numeric vectors to find content with similar meaning

Repository description (English)

An agentic LLM-powered knowledge assistant that enhances RAG capabilities through automated entity extraction, structured data analysis, and SQL-based reasoning to deliver accurate answers for complex, multi-step, and analytical queries.

Open on GitHub

Trending repos

All repos →

Latest from METAL MEDIA