MACHINE
A feature goes in, a verified pull request comes out.
AI-Driven Dev Engineer, Paris
I build at the speed of a team.
Parallelising the development of a single feature.
That is what MACHINE does, the harness I built. One feature, split into independent tasks run side by side, without skipping a specification, a test or a review.
- Based in
- Paris, France
- Since
- 2014
- Ground
- Web, native mobile, AI in production
- Format
- Product engineering assignment, freelance
- Company
- NOKEN
Where I have workedEmployers and client organisations, not customers of mine.
- Airbus
- Thales
- Capgemini
- SNCF
- Hero
- Expleo
- Meritis
- Assystem
- Eco-Emballages
- Pop Transport
The tooling
What explains the speed.
Two tools I built, and that I run in front of other engineers.
COCKPIT
Coding, made visual. The run, above the session.
The chain
/forge
- Idea
- Spec
/machine
- Spec
- Plan
- Architecture
- Parallelism
- Implementation + tests
- Adversarial agents
- Draft PR
Merging and deploying: human.
Three levels of parallelism
Both products, in detail- 1The tasksThe feature is partitioned into file-disjoint tasks. Up to eight run side by side, each on its own branch, in its own git worktree.
- 2The readingsInside each task the adversarial review splits into three simultaneous readings — OWASP security, logic and edge cases, clean code — and a fourth at senior grade: architecture conformance.
- 3The queueA last level chains whole runs over a closed queue, under a cycle and cost ceiling, until the target is closed.
Portfolio
What I have built.
What you can bring me in for
Four ways to work together.
Ship a product from zero
From the business need to deployment. The web, the native mobile app and the models, by the same person, with no seams between three vendors.
Take over a platform
An existing system to hold, stabilise and move forward, without rewriting it from scratch because the code is unfamiliar.
Industrialise a model
Model orchestration in production: cost per call, latency, retries on failure, and a reproducible result.
Speed up a team
Install the chain I use myself: framing, anchoring before the code, automated review, verified deployment. And leave it behind me.