Data engineer in the UK. I build small, on-prem tools that make it safe to point an AI agent at a real database or a real payment rail — and to audit what it did. MSc Data Analytics, Aston.
- sql-steward — the agent never writes SQL. Queries compile from a semantic layer; blocked PII is refused before the query runs.
- governed-agent-stack — the on-prem stack around it: map the schema, query it, audit every step.
- steward — delegated spending, for when the person spending and the person paying are not the same person.
- pay-warden — an MCP server that policy-checks a purchase before it reaches the payment rail.
- payoptimize — routes each payment to the provider most likely to authorise it. Thompson sampling, decline-aware cascade.
- sql-sop — rule-based SQL linter on PyPI. 48 rules, SARIF output, pre-commit hook.
Python · SQL · dbt · FastAPI · Ollama / LangGraph · Docker. On-prem by default.