# silou.dev — Silouane Galinou

> Agentic AI advisory, coaching & training by Silouane Galinou — senior
> full-stack dev & tech lead, Paris. I help teams that already pay for
> Claude Code get real production leverage from it, by building the
> harness around the model: context, agents, hooks, evals, observability.

Every page of this site has a markdown twin — the same content as the HTML,
no navigation chrome, with prices and booking URLs resolved to literal text.
Twins are generated at build time from the HTML source of truth. This file
is the index; each entry links a twin.

The site's UI is also published in French at https://silou.dev/fr/ — same content,
same structure. It has no markdown twins: the long-form writing there is
still the English original, so this index stays the single, English one.

## Pages

- [Home](https://silou.dev/index.md): Senior dev & tech lead. I help teams that already pay for Claude Code get real production leverage from it — by building the harness around the model, on your setup, in real time.
- [Sessions & pricing](https://silou.dev/sessions.md): Hands-on AI working sessions on your own Claude Code setup. Individual rates published; teams and companies, let's talk. Book a session.
- [Approach — how I work & how I earn your trust](https://silou.dev/approach.md): Practitioner, not influencer. I prove competence live on your screen (screen-share, no access needed), work my method in the open, and frame production experience honestly — no hype metrics, no name-dropping.
- [Lab — things I built that you can use](https://silou.dev/lab.md): Instruments and harness parts, published as I go: an open-source Claude Code observability instrument, MIT-licensed agent skills for Claude Code and Codex, a voice loop for coding agents, and an autonomous dev-workflow system. Honest WIP, not flagships.
- [Working notes](https://silou.dev/writing.md): Working notes on agentic engineering: the harness thesis, plan → build → review, LLM fundamentals, model-offloading, loop engineering, and two pieces that train a neural network and then take it apart.
- [Your AI Usage Card](https://silou.dev/pulse/index.md): A prompt you run in your own Claude Code. 100% local, ~2 minutes. It measures how you actually operate the model and prints a calibrated card: your level (L0–L5), your ≈ Cherny adoption step, and your one next unlock.

## Writing

- [Behavior over kit: what your Claude Code setup says about you](https://silou.dev/writing-behavior-over-kit.md): Installed tools tell you almost nothing — the level lives in behavior. The story of the AI Usage Card: a prompt that measures how you actually operate Claude Code, built to calibrate a talk for a room of ~50 engineers.
- [The harness, not the prompt](https://silou.dev/writing-harness-not-the-prompt.md): Why the next dev skill is engineering the system around the model — context, the agentic layer, hooks, offloading, observability — not wording the request.
- [What every team should know about LLMs](https://silou.dev/writing-llm-fundamentals.md): The mental model that stops cargo-culting: tokens, the context window, tools, and what an agent actually is — versus when you don't need one.
- [Model-offloading & the price of lost confidence](https://silou.dev/writing-model-offloading.md): When sending non-interactive work to a smaller or local model wins — and when the harness you need to trust its output costs more than the model you saved.
- [The same network, written out](https://silou.dev/writing-network-written-out.md): The mathematics of a digit classifier, drawn by hand: one pixel at a time, then sigma notation, ReLU, softmax and backpropagation — with the real activations from my own trained model.
- [Getting yourself out of the loop](https://silou.dev/writing-out-of-the-loop.md): Loop engineering plus scheduling (cron) turns a good harness into unattended, trustworthy runs — work that happens without you in the loop.
- [plan → build → review](https://silou.dev/writing-plan-build-review.md): The loop I work by: scope the work, let the agent build against that scope, then review — with evals so 'it works' is measured, not a vibe.
- [Train a network, then look inside it](https://silou.dev/writing-train-and-look-inside.md): An afternoon exercise: train a 109,386-parameter digit classifier in 82 seconds, then open it up and look at what it taught itself. Real figures from my own run.

## Other

- [About — who I am](https://silou.dev/about.md): Silouane “Silou” Galinou — agentic engineer in Paris. EFREI, eight years from junior dev to tech lead, in AI since 2023, eighteen months building the harness on Claude Code. Rides one wheel, climbs, reads everything.

## Contact

- Email: silouane.galinou.dev@gmail.com
- Book a session: https://cal.com/silouane-galinou
- Free 20-min intro: https://cal.com/silouane-galinou/intro
- LinkedIn: https://www.linkedin.com/in/silouane-galinou/
- GitHub: https://github.com/silouone