Embedded delivery
embedded customer delivery — SecondChair, a privilege-aware legal-AI system shipped to a small practice in seven days, in production
SecondChair is a retrieval system built for a small practice and delivered in a seven-day engagement, embedded with the practice rather than specified from outside. It runs in production and has supported a matter that reached the Iowa Supreme Court.
The stack: a case-management sync keeps the index current with the practice's own source of truth; privilege-aware BGE-M3 hybrid retrieval runs over a Qdrant vector store and a Neo4j graph (Cypher), with privilege filters enforcing what may surface for a given query; and MCP case-work tools put the practice's actions in front of Claude, one working surface instead of several systems. The SecondChair case study carries the full architecture — the multi-pass pipeline, what v1 taught v2, the court-records ingest cascade, and the query surface.
Client and matter details are confidential; this and the study describe architecture and engagement shape only.