Notes
Occasional writing on systems integration, payments, and AI implementation.
Three agents, one rule: the AI can only add care
After the leave agent, we built two more for the same car repair workshop: one for tools and equipment, one for job ads, applicants and onboarding. Hiring is where AI gets risky, legally and ethically. How the agents work together through Odoo, why the AI can only ever make them more careful, how we measure them against people's decisions, and what testing taught us.
An HR agent that knows when not to decide
We built an AI agent that handles leave requests for a vehicle repair workshop, on open-source Odoo, as a proof of concept for a car repair shop that needs it. The AI was the easy part. Deciding what it may decide, and where people stay in charge, was the real work. What we built, what we threw away, and what we'd tell anyone automating decisions about people.
Testing AI agents through change: write the tests once, swap everything else
Every model upgrade, re-index or prompt edit seems to need a new round of agent tests. It doesn't have to. A methodology, and a working reference implementation, for regression-testing AI agents that keeps the test suite fixed while the components underneath it change.
A reference implementation: AI voice and chat for a clinic contact center
We built this one end-to-end rather than just designing it — real phone calls, a real scheduling integration, and the production bugs a demo never shows you.
Inside a real-time voice AI pipeline: what orchestration actually does, in sequence
A technical breakdown of the Pipecat pipeline behind our voice assistant reference implementation — every stage, in order, and the production lessons that only show up on a real call.
AI implementation: what actually works, and what quietly kills projects
Challenges, opportunities, pitfalls, and best practices for organisations implementing AI — models, context engineering, agents, MCP, inference, and where training actually fits.
Sociomatica is moving into AI implementation
Alongside systems integration and Wero Ready, we're now helping organisations start their AI journey — architecture through to production-ready agents.