Application: twelve weeks, shipped and handed over clean.

Why this contract, why me

React with TypeScript and Python with FastAPI are both daily tools for me. polar-flow is a FastAPI service on PyPI with OAuth and a remote MCP server; coolify-mcp is TypeScript, 42 tools, 603 stars and 18.8k installs a month; TidyLinker is a two-sided marketplace with a React front end that went from an unrecoverable codebase to live in 4 months. I have shipped LLM features to production with evals and a cost ceiling, and I have run enterprise CI/CD: seven years leading platform engineering at Pfizer, deploys from hours to minutes across 100+ microservices.

Straight answers. My cloud depth and certifications are AWS; I am lighter on Google Cloud and would expect to be productive on your stack inside the first week. Auth I have built with OAuth 2.1. UK resident, no visa sponsorship needed.

How I would run twelve weeks

A fixed twelve weeks leaves no room for a slow start or a messy end. This is how I run short engagements.

1. Week one ends with something deployed. A thin slice through every layer (front end, API, auth, the model call, the pipeline) in production-shaped infrastructure by the first Friday, so every later week is adding to a working system rather than integrating at the end.
2. Evals before prompts. An eval set from real inputs and a pass mark agreed with the client before the LLM feature is tuned. For coolify-mcp that was a tool-selection suite with a red-team pass; for Bellwether, critic and prose-gate nodes that fail a draft that cannot cite its claims.
3. Cost ceiling first, model second. My hosted agents ran 10.1M tokens through 1,349 calls last month for $4 at list price, because the budget was designed before the model was chosen. The cheapest model that passes the evals wins.
4. Handover is a deliverable. Tests, a runbook and a release gate the client's team can run without me, written as the work goes and not in week twelve. On Rezzy that meant 0 to 59 tests and 22 tables cut to 10 before I left.

The numbers, with sources

Operating numbers, last 30 days, generated 2026-09-26. Sources per tile in profile.json; hover a tile for its source.