Experiments & field notes
Small tools built to test a question rather than to ship a product. Each one starts from a problem worth understanding; some become research, a few become more. Nothing here is for sale.
- 01prototype2026
Conformity & Procurement Response Pack
Can a vendor answer an EU enterprise AI procurement questionnaire in an afternoon instead of three weeks?
A structured generator that turns an AI product description into a defensible first-draft response pack for EU buyer security and AI-governance questionnaires — mapped to live AI Act obligations and framed strictly as an information tool.
Claude · Next.js · EU AI Act mapping
- 02research2026
Role-based Compliance Agents
What does a DPO or AI Officer actually delegate to a model, and where must a human stay in the loop?
A small family of role-scoped agents (DPO, AI Officer) that triage governance questions, draft register entries and flag escalation points — each with an explicit human-in-the-loop boundary rather than an answer machine.
Agents · Human-in-the-loop · Governance
- 03research2026
AI Act Enforcement Tracker
Can enforcement signals across member states be turned into a queryable, citable data layer?
A regulatory data feed that tracks guidance, national implementations and enforcement activity around the AI Act, exposed as a structured API for builders who need to reason about obligations programmatically.
Data API · Regulatory tracking
- 04archived2025
Clinical-to-Legal Translation
Can a model reliably turn a clinical record into the factual scaffolding a plaintiff attorney needs?
An early workflow tool that reads medical documentation and produces a structured, source-linked chronology — translating clinical language into the evidentiary form legal work depends on, with citations back to the record.
Document AI · Healthcare · Legal
These are research artefacts and information tools, not legal advice or services.