GeoRag

What it does

GeoRag is a compliance reasoning tool, accessed via Streamlit. It takes product artifacts such as PRD as input, analyzes them for potential compliance signals, retrieves relevant legal clauses from a Pinecone vector index, and runs a geo-specific compliance reasoner to determine whether implementation needs to vary across jurisdictions.

How we built it

We used LangChain as the framework, with Pinecone as the database.

Built With

langchain
pinecone
python

Submitted to

TikTok TechJam 2025

Created by

I worked on the legislation indexing and retrieval sections of the RAG pipeline, and connecting the feature parsing to my sections.

This was my first time working on a RAG project, and I learnt a lot about how LLMs work, and especially how best to optimize for RAG retrieval accuracy.

Ivan Koh
I worked on writing the initial logic to query the LLM given an input.

I worked on writing the signals, triggers, negation and logic of deciding whether or not to call the LLM

I worked on the graph database that stores the laws that are interconnected and ensure no unlimited cycles of relation and queries

Daniel Kwan
Shaun Tan
Gordon Hong