What I’ve built.
My own projects and demos. Client cases will be here, with permission.
Own project
Opportunity Radar
An agent system that watches 20+ sources twice a day, respects each one’s usage rights and writes a weekly opportunities briefing.
Python · GitHub Actions · Agents · 65 automated tests
See how it works →Own project
Agent-run second brain
The system I work with: agents file what I capture, propose tasks and execute them once I approve. A reviewer checks every result.
Claude Code · Subagents · Notion · Human review
See how it works →Demo · synthetic data
Document intake for an accounting firm
An agent sorts what clients send at quarter-end, files it, works out what’s missing and drafts the reminder. A person approves every send. Measured in 6 recorded runs.
Claude Haiku 5.5 / Sonnet 5.5 · Checks in code · Evals
See how it works →Demo · synthetic data
AI value report
Each month it answers the question that matters: is it worth it? Hours given back to the team, their value at INE labour cost and the model cost, with a source for every figure.
Evals · INE · Charts with sources
See how it works →Demo · synthetic data
Answering foreign buyers
An assistant replies straight away, in their language, to anyone who writes to an estate agency; it shows only listings that exist and fit the budget, and leaves a handover card for the agent when a person is needed. Measured on 15 synthetic leads.
gemini-3.6-flash · Guardrails in code · Evals · AI Act Art. 50
See how it works →