ARNAUD
GUIOVANNA.

AI PRODUCT BUILDER — END-TO-END

I'm a french AI Product Builder based in Normandy.

I close the full delivery loop on my own: product discovery, specs, agentic engineering at the SOTA (multi-agent coding, MCP servers, parallel subagents, eval harnesses), stakeholder rituals (copil / copro), Linear ops, deployment to bare-metal. Solo, end-to-end, on a scope a full dev team usually splits.

For 10 years, I've supported training organizations through their digital transformation. Today, same mission, scaled with end-to-end agentic development. See the open-source builds below.

END-TO-END · SPEC → DEPLOY · SOLO SINCE 2024

[ ↓ ]scroll

01PROJECTS

> Every project below is open source and shipped full solo, end-to-end. Same playbook on each: own the issues, write the specs, design the architecture, code with agents in the loop, review every change. The vertical changes (framework, application, AI tooling); the playbook does not.

OUVRIER

BUILDINGV0.1 · MIT

> A Go framework for agentic middleware. Declare a trigger, one or more LLM-powered pipes, the tools they can use, and the outcome — Ouvrier turns that declaration into a governed worker with typed results, tool sandboxing, traces, and durable state.

THREE PILLARS

runtime

From → Pipe → Reply. HTTP, Cron, Webhook, Stream triggers compile through one harness.

http · cron · webhook · stream

governance

Tool executor with permission policies, sandbox probe, hook bus. Every call is observable.

permissions · sandbox · hooks · traces

outcomes

Typed Output[T], JSON schema validation, durable state store, persisted event stream.

typed · schema · state · events

4

TRIGGERS

3

PROVIDERS

1

HARNESS

GO

1.25

MIT

LICENSE

Go 1.25 · Anthropic / OpenAI / Ollama adapters · SQLite state · sandboxed tool executor · Bubble Tea TUI

[ github ↗ ][ readme ↗ ]

TUTOR-MCP

ACTIVEV0.3 · MIT

> An adaptive learning engine exposed as an MCP server. It turns any LLM into a personalised tutor with real-time cognitive state tracking, spaced-repetition scheduling, misconception diagnosis, and a metacognitive loop that helps learners become autonomous.

THREE PARALLEL LOOPS

learning

What to learn, when. Five cognitive algorithms update the learner model in real time.

BKT · FSRS · IRT · PFA · KST

metacognitive

How the learner learns. Calibration tracking, autonomy score, mirror mechanism.

affect · calibration · autonomy · mirror

motivation

Why this matters. Signal-driven briefs, interest phases, no canned pep talk.

milestone · growth · reframe · utility

30

MCP TOOLS

5

ALGORITHMS

9

ALERT TYPES

90

TESTS

0-1

AUTONOMY

Go 1.25 · SQLite · MCP Go SDK · OAuth 2.1 · JWT · cron · Discord webhooks

[ website ↗ ][ github ↗ ][ readme ↗ ]

02STACK

LANGUAGES & FRAMEWORKS

  • Go
  • Next.js
  • Tailwind

AGENTIC CODING

  • Claude Code
  • Codex
  • OpenCode

LLMOPS

  • vLLM
  • LiteLLM
  • Langfuse

DEVOPS

  • Git
  • Docker
  • Linux

03SKILLS

> End-to-end means the four phases a team would usually split (product, stakeholders, engineering, ops) are handled by one person. The point isn't to do everything; it's that the seams between phases never become handoff costs.

PRODUCT & DELIVERY

  • spec

    discovery · PRDs · scoping · acceptance criteria

  • delivery

    roadmaps · milestones · scope cuts · async updates

  • linear

    cycles · triage · projects · sub-issues · ops

STAKEHOLDER OPS

  • rituals

    copil / copro · steerco · client demos · workshops

  • communication

    written status updates · decision logs · async-first

  • support

    GLPI · ticket triage · end-user comms · incidents

AGENTIC ENGINEERING

  • standards

    agents.md · memory.md · claude.md · hooks · skills

  • mcp

    server design · tool taxonomy · governance · evals

  • workflows

    TDD · subagents · parallel pipelines · persistent ctx

PLATFORM & DEPLOY

  • infra

    VPS · Docker · reverse proxy · TLS · cron

  • ci

    GitHub Actions · checks · release gates · rollbacks

  • observability

    logs · traces · alerts · post-mortems