THE AIGEN LAB
The AIGEN Lab  /  Los Angeles, California

Guillaume Tourneur

AI Product Architect — multi-agent systems, product, media, sports

I build multi-agent systems that change how products and media get made. AIGEN is the lab where the architecture gets proved. Everything below is live and clickable.

The last six months

15
AI products
in production
900+
Enterprise
users
20+
MCP servers
architected
28
Builds in
portfolio

The last four years — AI

40+
Agents designed,
built, deployed
11
Fine-tuned domain
agents in XCT
75%
Production cost cut
on a series
5
Episodes produced
through the pipeline

Ten-plus years — product

$20M
Product portfolio
managed
12+
Products shipped, all
met or beat KPIs
100+
Product enhancements
delivered
135K+
Users on a global
platform launch
01

The model

A lab I fund myself, so nothing here is a slide deck.

AIGEN is where I build. Not a consultancy pitch and not a portfolio of client logos — a working lab where I design multi-agent systems, ship them as real products under real brands, and find out what breaks before anyone is paying for it.

Three of those products are public and running right now: a multi-agent research-to-previz platform, a media company whose production team is a named roster of AI agents, and a science-fiction series produced through that platform. Anyone can open them.

The lab runs on a production stack as much as an AI stack, because the output has to survive a real audience, not just a demo. Twenty-six years in broadcast, media technology and product; the last four spent building the agents that build the work.

What the lab proves goes to work at scale. The same architecture — coordinator agents, subagents, MCP servers, an enterprise knowledge layer — is running in production today inside a global retail and lifestyle brand: 15 AI products live, 900+ users, 15+ business units. The lab proves it. The enterprise scales it.

02

Live work — open any of these

XCT design-brief schematic rendered on a desk surface, from The X Collective platform.
● Live product Multi-agent Deep search SORA previz

XCT v2 — The X Collective

Idea → cited research → design brief → previsualization

A multi-agent laboratory that turns a one-line idea into a defensible product concept. Deep-search agents mine journals, patents and the open web for a cited science-and-technology roadmap. Design agents convert those findings into a manufacturable concept and a structured design brief. Prompt agents render high-fidelity previsualization through SORA. Eleven fine-tuned domain agents underneath.

Used for world-building across gaming, film, product design and journalism — and used in production to make the series two cards down.

Stage 01
Deep-search agents
Journals, patents, web → cited roadmap
Stage 02
Design agents
Insight → manufacturable concept
Stage 03
Brief assembly
Structured, citable design brief
Stage 04
Prompt agents
SORA previsualization
Try it  → V2 release notes →
AIRWOLF contact sheet: expedition stills from THE WOLF — solar aircraft over dunes, field navigation, desert crossing.
● In production 10-agent system Media & entertainment

GBY — THE WOLF

Film company · production team designed by AIGEN

A production company making cinematic adventure documentaries about self-transcendence through science and technology. The distinguishing fact: its development team is a roster of named AI agents I designed — each with a discipline, a voice and a job in the pipeline, working alongside the human crew.

Two seasons in development: AIRWOLF, a solar-powered crossing of the Sahara, and SEAWOLF, ocean exploration built on biomimetic monofin engineering. The 10-agent production system cut production cost by 75% and time-to-market by 50%.

WERNER X
Writer, filmmaker
SOPHIA
Marketing, media, tech
BEN HAYASHI
Sponsorship, tech & VC
MANSA
Biologist, anthropologist
MAMORU
Environmental engineer
SEAWOLF contact sheet: freediving stills, biomimetic monofin engineering render, and ocean-going crew from THE WOLF.
THE WOLF → GBY → Watch →
Frame from The Humano Chronicles: helmeted figures in a dense future city, from the produced episodes.
● 5 episodes produced Built on XCT Anthology

The Humano Chronicles

Post-human tales · under ten minutes at a time

A hard science-fiction anthology — five episodes produced, each helmed by a different artist or scientist. It exists to prove one thing: the XCT pipeline survives contact with a real production schedule. Multi-agent scripting takes a concept from pitch to screen in days rather than months.

Latest episode: EPØ5 — Leif Attar, on humanity's final exodus and the seeding of life across the cosmos.

Watch the series → YouTube →
03

At enterprise scale

Same architecture, 900 people, seven months.

I stood up an enterprise AI capability from nothing — operating model, four-stage delivery framework, governance, portfolio, and data sovereignty through persistent storage and enterprise skills. Forward-deployed: I embed with the business unit on one-week sprints. POCs in days, products in weeks.

15
AI products
in production
900+
Active
users
15+
Business units
served
28
Build portfolio
under management
04

How it's built

Coordinator agents, subagents, and a knowledge layer that belongs to you.

No single mega-prompt. A coordinator decomposes the task, routes it to subagents that each own a narrow domain, and holds the agentic loop — tool use, escalation, human-in-the-loop where the stakes require it.

Architecture
Coordinator agents and subagents, task decomposition, orchestration, agentic loops, context engineering, tool use, escalation and human-in-the-loop
Models & tooling
Claude, Claude Code, OpenAI, Gemini, Mistral, Gemma · Python, prompt engineering, JSON schema, structured output, fine-tuning
Connectivity
Model Context Protocol (MCP) — 20+ servers architected, connector architecture, enterprise knowledge layer
Institutional IP
Skills libraries encoding business rules and organizational knowledge into agent reasoning, so the company owns the thinking, not the vendor
Retrieval
Retrieval-augmented generation, vector databases, knowledge graphs, NLP
Delivery
Four stages — POC → MVP → V1 → Enterprise, on one-week sprints with standardized intake, acceptance gates, evaluation criteria and architecture review before production
Compliance
HIPAA, HL7, FHIR, ISO, GDPR — from digital health and regulated deployments
05

Published

AI in Sports: How Wearable Data & AI Can Transform Boxing Performance

160 pages · March 2025 · with POWA

A full-length research paper on integrating biometric wearable data with reasoning models for combat-sport performance. Written with POWA; the applied version of the same work moved measured athlete performance by 25%.

Read the paper →

Wearables, Machine Learning & Endurance Sports

Presentation · 2019

Where this started. Machine learning applied to wearable sensor data seven years before it became a job title — the through-line from Fortune 100 wearable pilots to today's agent architecture.

Open the deck →
06

Field notes — what actually holds up

  • Adoption is the product.
    A model that works and nobody opens is worth zero. The 900 users came from AI champions inside each business unit, hackathons, a 23-module learning hub and a bi-weekly showcase running at 200+ attendance — not from the model choice.
  • Ship in one-week sprints or don't start.
    Long discovery kills enterprise AI. Embed with the business unit, POC in days, MVP in weeks, and let the acceptance gate — not a steering committee — decide what survives.
  • Sovereignty beats cleverness.
    The durable asset is not the prompt. It is the knowledge layer and the skills library encoding your business rules — portable across models, and yours when the vendor landscape shifts again.
  • Decompose, don't inflate.
    One coordinator plus narrow subagents beats one enormous agent every time. Smaller context, testable behavior, failures you can actually locate.
  • Build it on yourself first.
    Every architecture on this page was run against my own projects, my own money and my own deadlines before it went anywhere near an enterprise budget.
07

Track

Wearable sensors in 2014. Agent architecture now.

Product and technical program leadership since 2000 — broadcast, media technology, enterprise computing, two co-founded startups. The line into AI starts with putting models on sensor data, and it has not broken since.

2014–2018
G Advisory — wearable technology for Fortune 100. First piezoelectric wearable pilot for a top-three sneaker brand; mobile health platform for 40,000 RiteAid employees.
2018–2021
Movember Foundation — launched the first global digital health platform, 135,000+ users, HIPAA and HL7/FHIR compliant.
2019
Published Wearables, Machine Learning & Endurance Sports.
2020
MIT Professional Education — Machine Learning: From Data to Decisions. Fire Hydrant Award.
2022–2023
OutcomeMD — Director of Product Management, AI-driven digital health. Key contributor to a $3.7M convertible note; six months as acting CTO.
2023–2025
AIGEN — the lab opens. Multi-agent systems across media, healthcare, sports and private equity. GBY, XCT, SocialDream, POLAR and POWA.
2025
160-page research paper on wearable data, AI and boxing performance with POWA.
2026
Global retail and lifestyle brand — enterprise AI program leadership, reporting to the SVP of Data and AI. Zero to 15 products in production.

Field-tested on myself — boxing, Ironman, expeditions. @guillaume.tourneur

08

Contact

Looking for a Director or VP seat where the AI actually ships.

Los Angeles. US and France dual citizenship, authorized to work in the United States without sponsorship. No forms — write to me directly.

guillaume@aigenconsulting.com LinkedIn →