I work with founders, CTOs and boards on the technical decisions that come before a significant investment. I bring architectural discipline to AI-assisted development.
You have an MVP to build, starting from an idea or a prototype, and need senior technical leadership to design and deliver it.
Your system starts to break under growth: response times, security findings or infrastructure costs grow faster than the number of users.
Your software is built by a supplier and you need someone on your side who can judge estimates, answers and releases.
After ACI group acquired its traffic agency, I assessed the inherited PHP and Oracle system: it could not scale and went down repeatedly. I designed its replacement: 18 services on AWS that merge six sources on DATEX II, GTFS and TMC every 3 to 5 minutes and serve about 10,000 events a day to HERE, TomTom, Radio24 and others.
54% of database waits came from writing a log row for every API request: 2 billion rows, 13 TB. Caching token validation cut total SQL time by 77% and database CPU by 89% in load tests. I led the architecture review and the caching strategy with Redis on top of the database.
12 servers processed 6,000 files an hour, but searches took minutes and history was limited to two months. MongoDB won a competitive PoC against SQL Server and Oracle. In the design benchmark, the new write path sustained 537 writes per second against the 300 required for ten times the volume.
A behavioural assessment method, which scores candidates against role profiles, lived in a spreadsheet with its algorithms and 7,500 past responses. I rewrote the scoring engine in Java on RESTHeart and MongoDB, with tests that compare profiles and population statistics against the spreadsheet. The multi-tenant web application reached production in under three months.
A single shared e-commerce platform on Adobe AEM for a luxury brand: frontend in Berlin, backend in Gurgaon, deployment in Shanghai, client in Geneva. Local compliance and Chinese localization were built into the design.
I co-created RESTHeart, the open source backend for MongoDB, in 2014: more than ten years in production, adopted by Unisys, Northrop Grumman, Crownpeak and others. Ulabase, MobiStream and Sophia, a RAG platform, are built on it.
Building software with AI keeps getting faster. When producing code takes less time, it matters more to know what that code will cost in fixes, maintenance and downtime.
AI-written code almost always looks good, even when it is fragile inside. To assess it, you need to know the practices by which it was designed, tested and prepared for production.
I organize agentic development with Lean engineering practices.
Agents need explicit job descriptions and hard guardrails. The workflow operates as a pull system: well-specified tasks wait on a board, where dedicated agents pull work within their scope, implement changes, and validate them against tests.
Without that structure, agents generate more code than anyone can review.
A new product starts with a Discovery. It takes two to four weeks at a fixed price.
I start from the goals of the initiative, the obstacles in the way and the criteria that measure success. I map the main processes and the people involved.
I choose the architecture, the technologies and the infrastructure of the system. If a prototype exists, I use it as the specification. I keep its technologies when there is a reason to.
I define what the project includes and what it excludes, and in which order. Each possible approach comes with its assumptions, dependencies and risks, and an indicative timeline.
You get a written document with enough information to decide go or no-go, and a realistic cost range for development. It is yours and can be used with another supplier.
Tell me what you're building.
