System 01 — online · Martin Bouvet

I build AI systems that work while I don't.

I'm a business student at emlyon who studied LLMs, generative and agentic AI at Oxford. I treat AI agents as a team: they capture what I learn, consolidate it at night, write code with me and ship products. This page is the proof.

See the work ↓ GitHub ↗
    12public repositories
    605tests in public repos
    passingCI on nightshift
    todaylast push to a repo
    Repositories, CI and last push are read live from the GitHub API when this page loads; the test count is declared, from each suite. The same panel as my GitHub profile header, which rebuilds itself every morning.
    Selected work

    Five case studies

    Two agent systems, two products, and one venture taken from market thesis to a working v0. Each page shows the problem, the architecture, the decisions and honest numbers.

    01 AI product · live

    Cortex: course handout in, study system out

    Drop in a PDF, get a study sheet, quizzes, spaced repetition and a tutor that only answers from your course. Freemium with Stripe, admin panel you can run from a phone.

    liveNext.js 16ClaudeStripe33 tests
    Read the case study →     02 Agent system · daily

    The AI OS: a second brain with a night shift

    Saved videos become scored notes twice a day; an agent consolidates them at 00:30; my other agents search it before answering.

    open sourcePython 8.3k linesClaude CodeObsidian
    Read the case study →     03 Agent system · prototype

    HAO: agents with a chain of command

    LLM agents organised like a company. They decide, use tools, and escalate to their manager when their confidence is too low.

    TypeScript3 LLM providers55 tests
    Read the case study →     04 Product · live · open source

    Nous Deux: warm product, paranoid backend

    A private space for long-distance couples. Row-level security on every table, content-free notifications, 48-hour location retention.

    liveReact 19Supabase430 tests
    Read the case study →     05 Business × product · live

    Diavie: regulation read before code written

    A French companion for people living with diabetes. Mapped against EU medical-device rules first, local-first by design, gated by evidence.

    live€0 to runPWA
    Read the case study →
    From market thesis to working product

    The business side

    Code is half the job. The other half is finding who pays, for what, and packaging it. No paying customers yet; here is what's on the table.

    Active offer

    E-invoicing readiness audit

    France is rolling out mandatory e-invoicing: receiving from September 2026, issuing from September 2027. I'm packaging a fixed-price audit that gets small shops and craftspeople in Lyon ready before penalties apply.

    Productised service

    AI Second Brain kit

    The system behind my own AI OS, turned into a deliverable: a vault template, a timed delivery checklist and a client guide, priced as a one-off service.

    Opportunity scouting

    Business Idea Radar

    Scans Reddit for recurring pain points and ranks them as business opportunities, with an optional Claude check. How I look for problems worth solving.

    Also built

    More projects

    About

    Business student by day, builder by night. I care about one question: what does this create, for whom, and is it worth paying for?

    1. Studied LLMs, generative and agentic AI.
    2. First commit. Started shipping with AI agents.
    3. Nous Deux, Cortex, Diavie and the AI OS go live.
    4. Global BBA at emlyon business school, Lyon. Launching the e-invoicing audit.
    5. Open to conversations about AI automation for small businesses.