Due Diligence Risk Agent

Problem

Risk reviews combine narrative answers, judgement, and repeatable scoring. The workflow needed a clearer shape before any model could be trusted with it.

My role

Designed the workflow and risk model around impact, likelihood, score, and rationale, then translated it into an AI-ready integration pattern.

What was built

A structured assessment flow that gives reviewers a consistent place to inspect the inputs, reason about risk, and retain the rationale behind a decision.

Architecture

questionnaire inputs → structured risk fields → model-assisted analysis → reviewer decision

The client wired the integration to its own model through scripts.

Evaluation & controls

The important control is the structured review step: impact and likelihood remain visible, and the model does not silently replace the reviewer’s judgement.

Trade-offs

Keep the first version close to the existing questionnaire and decision language rather than introducing a broad autonomous agent with unbounded scope.

Evidence of completion

Local portfolio evidence describes the workflow as designed and prototyped, with client-validated time savings. The underlying client materials are not published here.

Next step

Turn the prototype into a redacted, synthetic case study with a small evaluation set and an explicit approval record.