All projects
AI / LEGAL·LIVE · 2026
Nova Intelligence Engine
A multi-model AI platform indexing 40M documents for a global law firm. Grounded answers, full citations, private inference.
OVERVIEW
A global law firm needed associates to search forty million documents without ever risking a hallucinated citation. Nova indexes the full corpus, grounds every answer in retrieved source text, and runs entirely on private inference infrastructure the firm controls end to end.
THE CHALLENGE
Legal research demands a citation for every claim, zero data leaving the firm's infrastructure, and answers fast enough that associates actually use it instead of manual search.
OUR APPROACH
- 1Built a hybrid retrieval pipeline over a 40M-document Weaviate index
- 2Constrained generation to always cite the retrieved passage it draws from
- 3Deployed private inference on a dedicated GPU cluster, no third-party APIs
- 4Ran a six-week shadow evaluation against senior associates before rollout
RESULTS
40M
Documents indexed
100%
Answers with citations
0
Data leaving the firm
TECHNOLOGIES
PythonLangChainWeaviateGPU cluster
NEXT CASE STUDY