Skip to content

AI & LLM

Legal Assistant AI

Enterprise RAG for legal research

Role
Full Stack & AI Engineer
Year
2025
Status
Case study
Stack
5 technologies
Read the LinkedIn post
Legal Assistant AI — Enterprise RAG for legal research

  • 94%

    Answer accuracy

  • 70%

    Faster research

  • 5+

    Law firms

Screens

03 · click to zoom

Overview

Lawyers ask questions in plain language; the assistant embeds the query, retrieves the most relevant passages from a ChromaDB vector store and has GPT-4 answer strictly from that context.

The result is research that is faster and traceable — every answer comes with the sources it was built from.

What I built

  1. 01Retrieval pipeline: embedding model → ChromaDB vector store → top-k retrieval → GPT-4 with grounded context.
  2. 02Answers cite the exact section and source file (e.g. Section 148(g) of the Legal Profession Act 1966) and show a confidence level plus the top source chunks with similarity scores.
  3. 03Document library: upload legal PDFs, automatic chunking (141 chunks for a single Act), processing status, search and type filters.
  4. 04Chat history with search, favourites and filters; save and export conversations.
  5. 05Built-in legal disclaimers on every response and role-based accounts for lawyers.
  6. 0694% answer accuracy on legal queries.
  7. 0770% faster legal research for the teams using it.
  8. 08Deployed at 5+ law firms, saving an estimated $50K+ per firm annually.