← All work

Legal researchBeta

LegaFetch

A legal research engine that searches by meaning instead of keywords, returns the passages it found and writes an answer that cites them. It works from the web UI and through an API.

Scope
Product, Retrieval pipeline, Infrastructure
Stack
TypeScript, Python, Qdrant
LegaFetch — home page of the live product

The problem

Finding precedent means guessing the exact words a judgment used. Searching by concept removes the guessing; showing the source passage under every claim keeps the answer checkable.

What we built

  • Semantic search over Yargıtay, Danıştay and Constitutional Court decisions plus current statutes
  • Cited answer generated only from the retrieved passages, with sources listed
  • Document upload: topics are extracted and matched to legislation; the document is not stored
  • Licence-key access and a public API

Engineering notes

Say what the engine is not

The page states the limits: an LLM writes the answer so citations must be checked, it does not give legal opinions, and statute coverage is limited to 73 laws.

By the numbers

9,860,914
court-decision passages
15,479
statute passages (73 laws)

legafetch.com, measured 20 Sep 2026