Skip to main content
Back to selected projects

Applied AI projects

Six smaller cases, each built around one product decision.

The cases cover contract intake, contract review, interview scoring, policy review, API packaging, and agent/browser controls. Interactive cases use fictional or synthetic data. Cases that cannot be released safely stay static.

Internal-pilot origins · Clean-room public treatment · No private repositories, production data, or adoption claims

Project library

Why each case uses a different public format.

The format follows what can be shown safely: an interactive demo, guided replay, technical case, static case, or architecture note. These are supporting cases, not six additional flagship products.

Global Contract Preapproval

An interactive reconstruction of contract preapproval: parse a fictional intake, show where each field came from, run fixed risk rules, and record a human decision.

Interactive Demo · Clean-room reconstruction · Synthetic data

Field set, risk states, human decision, synthetic workbook, and action timeline

Open evidence

Contract Review & Redline

A fictional contract review that keeps rule matches, prepared AI suggestions, and the reviewer's accept, modify, or reject decision separate in the audit trail.

Guided Demo · Clean-room reconstruction · Fictional contract

Finding trail, prepared suggestion, human decision, and printable HTML redline

Open evidence

AI Outcomes Interview

A fictional interview replay with six prepared scores. The reviewer can change each score, write feedback, and save an audit record; the test record never enters ranking.

Product Replay · Prepared fictional transcript · No real ranking data

Six-dimension scorecard, human adjustments, feedback, and audit record

Open evidence

Policy Review Copilot

A static walkthrough of fictional policy clauses, the issue found in each one, the cached support, a prepared recommendation, and the reviewer's final decision.

Static Case · Cached and self-authored evidence only

Issue, evidence, recommendation, and human-decision trail

Open evidence

API-to-Agent Forge

A technical case that maps one fictional Spring Boot controller into four prepared outputs: OpenAPI, CLI help, an Agent Skill, and a validation manifest.

Technical Case · Prepared outputs · No public generator

Prepared OpenAPI, CLI help, Skill example, and validation-manifest downloads

Open evidence

Multi-Agent Workbench

A static architecture note showing how agent, browser, relay, and server layers are separated, where human confirmation is required, and why no executable demo is public.

Architecture Note · Non-executable public treatment

Static architecture and non-release rationale

Open evidence