Knowledge Extraction Engine

The Problem

A senior professional with 10+ years of experience has hundreds of rules in their head — when to push, when to wait, what red flags to watch for. That knowledge is the most valuable asset in the business, and it only exists in one person's brain. If they leave, it walks out the door.

The Approach

I built a system that extracts this knowledge systematically. An AI interviewer conducts structured sessions using Critical Decision Method — walking the expert through real scenarios, identifying decision points, and extracting the underlying rules in real-time. Each rule is structured: situation triggers, required actions, reasoning, exceptions, numerical thresholds.

Key Technical Decisions

1. 1.Jaccard similarity matching: prevents duplicate rules across interview sessions. The system catches when the expert says the same thing two different ways.
2. 2.Exception detection: auto-identifies when rule data is in the wrong field. Found 28 misclassified entries out of 279 rules — the system flagged every one.
3. 3.Progressive disclosure: rules are organized into 3 levels of detail. AI agents downstream load only what they need, keeping context windows tight.
4. 4.Publishing pipeline: review, confirm, publish with version tracking. Generates both markdown and JSON for downstream consumption.

Results

Stack