Knowledge Finder

A per-team knowledge graph answering questions over contracts, reports, and email with page-level citations.

Client
Private client
Tag
AI Retrieval
Knowledge Finder screenshot
Knowledge Finder project image 1

Problem

Thousands of contracts, reports, and emails sat across shared drives and inboxes, forcing manual searches for answers like upcoming supplier renewals. Keyword search returned files, not answers, and plain chatbots lacked proof. Legal, Finance, and Operations needed isolated graphs on client infrastructure, with strict access controls and every claim cited to a source page.

Knowledge Finder project image 2
Knowledge Finder project image 3

What we built

A five-part graph retrieval system delivered over 2 months. Daily ingestion pulls drives and inboxes; an LLM extracts parties, dates, amounts, and clauses into one knowledge graph per team. Retrieval walks graph relationships instead of matching keywords and cites every claim to its source page, accessible via web app, API, and MCP inside the team's existing assistant.

Knowledge Finder project image 4

Outcome

Questions that once meant opening dozens of files return answers in minutes, showing both the graph path and underlying source pages. Legal, Finance, and Operations each query only their permitted graphs. At handover, a part-by-part walkthrough left the client's team ready to operate and tune the system.

Knowledge Finder project image 5
Next project
man | made