Design. Build. Ship.
Engineered for production.
Full-stack engineer, 9 years, end to end: back ends, front ends, data and delivery. I led the rebuild of a legacy platform used by clients across Asia and Europe without interrupting a single one. Today I lead architecture for Nooko.ai at EGS, a Swiss company, where I bring AI into a product people pay for.
Full-stack delivery for an AI culinary platform: Python and FastAPI services, React front ends, and AI workflows that act on real data safely.
Pydantic checks every call before it reaches the app.
A second model reviews and repairs bad output.
Iteration caps, token budgets, confirm before writes.
Tool-call success, spend and time measured per run.
Replaced a slow, vulnerable legacy system with microservices on Azure and AWS, while clients kept working every day.
Proposed and started it alone, then led 4 developers.
Old and new in parallel, clients verified their data, moved in waves.
Domain services behind an API gateway, RabbitMQ for long tasks.
CI/CD and Docker, releases with teams in Asia and Europe.
Two systems I led, in depth. Then client and side projects, and the free tools I build and run on this site.
An AI culinary platform for chefs and food brands. I lead its architecture and build the workflows that let models act on real data safely.
Models make things up, loop and overspend. None of that is acceptable when customers pay per result.
Architecture lead. Python and FastAPI services, React and TypeScript front ends, delivery on Azure.
Typed tool calls, a second-model correction loop, iteration caps, token budgets, preview before saving.
Failed runs refund credits automatically. Every run is measured on success, spend and time.
A slow, buggy legacy system on outdated, vulnerable software that could not integrate with anything. Customers used it every day.
It could not simply be switched off. Every client had to keep working through the move.
Proposed and started it alone, then led 4 developers for 2.5 years.
Old and new ran in parallel. Core features first, read-only migrations clients verified, cutover in waves.
About 27 clients moved, each with consent, with service never interrupted.
The first two come from my Full-Stack Developer years at CalcMenu, before the Cloud rebuild.
Voice commands in a flagship enterprise product, so users can work hands-free.
Automated attendance tracking with facial recognition.
Backend API streaming live data to dashboards.
Online store with inventory management, payments and an admin dashboard.
This site: static Astro front end with a Fastify API for comments, Google sign-in and reCAPTCHA.
Live on this site. No ads, no tracking, no sign-up.
Notes from the job: architecture, Python, AI in production and how I work. A new piece every two weeks.
NewMCP made it easy to describe a tool once and let the model call it; making those tool calls hold up in production was a separate job, and that is where most of my time went.
Read the article →NewMCP made it easy to describe a tool once and let the model call it; making those tool calls hold up in production was a separate job, and that is where most of my time went.
Practical lessons from deploying AI agents with tool-calling. Hallucinated arguments, infinite loops, runaway costs, and the guardrails that actually prevent them.
Production-focused guide to building RAG pipelines with Python. Chunking strategies, vector search with pgvector, re-ranking, prompt construction, and the stuff that breaks past the tutorial stage.
My AI coding bill reached about $650 a month on GitHub Copilot. I spent a month on a free local model, went back to Claude, and now use both.
Python patterns I rely on for production microservices: structured config, graceful shutdown, real health checks, idempotent message processing, and structured logging. All with code.
A senior engineer's honest take on AI-assisted development. What actually helps, what's a waste of time, and how to keep your skills sharp while shipping faster.
How I led the rebuild of a slow, buggy legacy platform into a microservice-based one without interrupting customer service, and the unglamorous work that made it possible.
Nothing matches that yet. Try another word, or .
Architecture lead for Nooko.ai, an AI culinary platform, from requirements to production.
Python · FastAPI · Pydantic · MCP · React · TypeScript · Azure
Python services that deliver client data files reliably.
Python · RabbitMQ · SQL · REST
Proposed and led the CalcMenu Cloud rebuild, alone at first, then with 4 developers.
ASP.NET · Python · Angular · SQL Server · RabbitMQ · Docker · Azure · AWS
Features and automation for the flagship product.
Angular · ASP.NET · Web Speech API · Python · OpenCV
Short full-stack projects and maintenance for several clients, often in parallel.
React · Angular · Python · SQL Server
Trinity University of Asia