Skip to content
Independent engineering studio · Paris / Berlin / Remote

Systems that hold up in production. AI included.

Architecture, backend and AI. I design, build and take over software systems for teams that already have users. Fifteen years of code, ten years leading engineering teams.

One request, several providers, one validated answer
Request, orchestration, interchangeable providers, validated result

No provider is a single point of failure. If one degrades, the next takes over, and nothing reaches your users before it has been validated.

Architecture & backend

Simple foundations, documented, ready to evolve.

AI in production

Interchangeable providers, validated outputs, tracked costs.

Taking over existing systems

Find the real cause, fix it cleanly, without breaking anything.

Recent work: the problem, what was built, the stack.

Digital training platform

client not named
Problem

An AI tied to a single provider, answers that were sometimes unusable, and a language bug running across several stacks.

Built

Multi-provider orchestration by capability with fallback, output validation with automatic correction, end-to-end fix across Kotlin, NestJS and the front end.

Stack

NestJS · Kotlin · GraphQL · OpenAI · Gemini · Mistral · Anthropic

What I build on my own account

Two products where the architecture, the AI orchestration and the billing get tested at my own expense before anyone else pays for them.

Mestri

Pilot underway · launch 2027

The diagnostic know-how of a technical SME, encoded once and available to the whole team. Multi-tenant, one vector database per client, European hosting.

Transcription

In development

Audio transcription billed by the minute of audio. Duration and cost shown side by side before you confirm. Retention visible, deletion in one gesture.

How the work runs

Scope written on day one

Scope, timeline and responsibilities on paper before the first commit. Anything outside gets discussed, not absorbed.

Visible progress

Written updates on a fixed rhythm, clear status, blockers raised early rather than at the deadline.

Ownership until handover

The outcome is carried end to end: quality, delivery, handover. No half-finished branches left behind.

Questions, answered

What exactly does CodiumLab do?

An independent engineering studio: architecture, backend, AI in production, and taking over systems that already exist. One senior engineer, accountable for the outcome.

Why a monthly engagement rather than a pack?

A pack ends. A live system needs capacity held month after month, not a fresh negotiation every quarter.

How is the scope handled?

Agreed month by month, with a written weekly review. What falls outside the month gets discussed instead of absorbed.

Remote or on site?

Paris / Berlin / Remote, written updates by default. On-site in Paris or Berlin now and then, not three days a week.

How do we start?

A 30-minute call. You leave it with a recommended entry point, or with a clear no.

Thirty minutes to find out if this fits.

Bring the system, the constraint and the deadline. You leave with a recommended entry point, or with a clear no.

Book a 30-min call