pi-context-tree
An extension for the Pi coding agent that injects the right context at the right time, by path and operation, from CONTEXT.json files versioned in the repository.
12 commits / 12 months
v0.4.2
ZEDIUM-Off/pi-context-treeWho I am
An engineer who stays until it is used
I learned the trade by shipping. Trained at 42 Lyon, I signed my first developer contract in 2022. Since then I have built systems that serve real users, mostly from inside my clients' teams: an after-sales automation in production in healthcare, a mobile app and its back office, a document-reading benchmark under health-data constraints.
In January 2026 I founded MC Studio to carry this work under my own name. My field: operational AI and agentic engineering, where security, traceability and adoption matter as much as the model. I work with agents every day, framed by written rules, skills and verified workflows; this site is built that way. That is the method I bring to the teams I join, and when a project calls for it, I can bring together a small team of hand-picked independents.
- First developer contract
- 2022
- MC Studio, founded in January
- 2026
- Trained at
- 42 Lyon
- Working languages
- FR · EN
Landmarks
A few facts you can check on this site
- client organisations, directly or alongside partners
5 - published case studies, limits included
9 - projects that integrate AI or train people to use it
7 - tools published on GitHub for working with agents, 5 of them open source
6
Every case study and tool counted here can be read on this site. I have shipped software since 2022, and MC Studio has existed since January 2026.
Expertise
What the market asks me for first
Operational AI and its adoption by teams come first: that is where projects most often fail, and where I am most useful. The rest of the chain follows, because an agent without reliable data or an application to live in is useless.
- Flagship role
Operational AI
When AI was tested without a criterion or a frame, and never left the pilot.
LLMs, agents, RAG, OCR and vision models, open-weight or self-hosted models: used where they bring measurable value, with clear data boundaries and human validation.
- Flagship role
AI adoption and training
When everyone tinkers with their own assistant, with no shared rule or measurement.
Taking AI from isolated trials to a team practice: usage rules, guardrails, agentic engineering training for developers, internal champions and usage measurement.
Data and automation
When the information exists but never reaches the right place.
Custom business applications
When spreadsheets, email and re-keying stand in for a system.
Architecture, environments and hosting
When technical debt and hosting slow down every change.
Working inside teams
I join your team, then I upgrade the way it works
I mostly work inside my clients' teams, on projects I did not all start. It gave me a method: join your rituals first, understand what already works, then set up, sized to the team, the agentic work primitives I use every day at MC Studio.
Rules written for humans and agents
An AGENTS.md, one page per project, recorded decisions. The newcomer and the agent read the same source, and it gets reviewed like code.
Know-how as skills
A recurring procedure becomes a skill the agent loads at the right moment. In my own work, preparing the monthly activity report is one, with its verification rules.
Graph workflows, with human approval
Long tasks become graphs of steps, with resumption and human decision points. That is what pi-graph and Zedflow are for, two tools I publish.
Proof before trust
Tests, checks and reviews are part of the flow. An agent that cannot show what it verified is not done.
Work
Proof files, not a gallery
Each case study reads in one go: the key points in thirty seconds, then the story, the proof and the limits.
Automating after-sales returns to manufacturers, with a trace of every request
Santécie’s after-sales return files are routed to the right manufacturer forms and filled automatically, with an audit trail for every request.
Training developers in agentic engineering, in eight sessions
I am preparing an eight-session path, designed to take developers from occasional use of an assistant to verifiable agentic engineering, including knowing when to do without it; first session on 16 October 2026.
Zedflow, agent flows you can read, compose and control
With Zedflow, an agent flow is read, validated and composed like a program, and sensitive decisions wait for a human approval.
Clients and tools
Who I work with, and with what
Clients and partners
Every logo links to the official site of the organisation or project. Trademarks belong to their owners.
For IT firms and agencies
My expertise, serving your clients
Your clients ask you for AI and agents, and your teams already ship a lot. I know subcontracting from the inside: your client stays yours, I bring the skills you lack, and I train your teams so they keep them.
- I do not take your territory
- I bring what you lack
- I train your developers
Agents and open source
Tools for engineers who code with agents
I publish the tools I build for my own work with agents: an agent-flow engine, extensions for the Pi coding agent, skills, my context configuration. Each with its install command, what it does and when to reach for it.
pi-runtime-monitor
A lightweight Pi extension that shows, in the footer, the CPU, memory and aggregated usage of your session’s subagents.
3 commits / 12 months
v0.1.1
ZEDIUM-Off/pi-runtime-monitorplan-writer
My shared agent skills, installable with the skills CLI; for now plan-writer, which helps write structured implementation plans.
4 commits / 12 months
ZEDIUM-Off/skillsArticles
Field notes
What I learn by shipping and by working with agents: methods, lessons, limits.
Joining a team, then upgrading the way it works with agents
How I join an existing team and install, sized to it, agentic work primitives: written context, recorded decisions, skills, workflows with human approval, proof before trust.
Automating a sensitive process without losing traceability
Automating a process that touches health data, payroll or invoicing is not about speed. It is about memory: knowing, afterwards, who decided what, under which rules and on which data.
Let’s talk about what AI could change for you
A 30-minute discovery call, by video and with no commitment, to talk about your project, your constraints and the best next step.