I turn voice notes, documents and evidence into structured, reviewable, decision-ready systems — not chatbots.
Expert work rarely starts clean. I map the mess — inputs, decisions, exceptions, handoffs — into systems that know what to extract, what to draft, where they're unsure, and where a human must decide.
Pre-accounting memory for sole traders
Compliance reports from site evidence
Private proof records for property compliance
Project delivery in complex built environments
Actors, inputs, decisions, exceptions, handoffs — how the work really happens.
Report, record, export or decision. AI is only useful when the output is clear.
Prompts, extraction rules, confidence signals, review points, escalation paths.
Prototypes and Codex/Claude implementation briefs engineering can execute.
Missing evidence, hallucination, weak confidence, polished-but-useless output.
The SlipMind demo is public on GitHub → Other repos are private — commercial workflows and compliance logic — but I share walkthroughs, architecture notes and redacted examples during interview.
For AI workflows in evidence-heavy, regulated or operationally messy environments.