Ask anything about Simon's work, experience or background. Answers are generated live by Gemini 3 Flash, using only this page.
Research Contributor at MIT Critical Data
Simon Heilles.
I build reliable AI: systems that survive contact with production. Building with LLMs since GPT-2 was published in 2019.
Book a call with me ↗LinkedIn 2,832 followers · Instagram 1,098 · follow me on X and GitHub
“If I have seen further, it is by standing on the shoulders of Giants.”
Now
I founded DENEM Labs, an applied AI lab and an official Anthropic partner. We help enterprises put AI into production and keep it trustworthy there, often on their own hardware when the data can't leave the building.
In practice that means RAG company-brains, automation agents, embedded-AI SaaS, and a private on-prem inference server. I run AI trainings and ship products for Veolia, Van Cleef & Arpels, Canal+ and Cofidis, and moderated an industry round table with an HR leader from Canal+. I'm also a member of the Google Developer Program. Off the clock, I volunteer with the Red Cross as a first responder and first-aid trainer, including at the Paris 2024 Olympics.
Long before any of it: an astrophysics internship at the Observatoire de Paris at 14, and a clothing label at 18 that I grew to roughly 3,000 followers.
Selected
work
Thomas (Y-Combinator P26)
Built an agent learning system that turns past experiences into reusable lessons, plus the benchmark that measures whether it truly learns across sessions.
CommonTrace
Agentic-memory harness: a repository of reusable traces, shared across sessions and across agents.
ESM-2 variant effect
Reproduced Meta's ESM-2 variant-effect-prediction pipeline from the research code and improved on their best baseline (ensemble stacking), with a GridSearch across five protein families.
VictorIA
A conversational AI assistant built for Veolia, guiding elected officials and citizens toward concrete local solutions.
Pierre AI
An internal knowledge layer that makes a company AI-native: ingests artifacts (Gmail, GitHub, Notion, Stripe, Meta Ads) into a queryable memory, with autonomous actions.
Health-AI CIR dossier
At 21, authored the technical section of a 2020 R&D tax-credit (CIR) dossier for a health-AI company (NLP for hospital coding), on contract via Research to Business: the state of the art, open problems and contributions of an industrial BERT R&D effort.
What it actually takes to put AI into production, from the DENEM Labs feed.
Experience
-
2026 →
Research Contributor · MIT Critical Data
Working at the frontier of AI for scientific research, with a focus on automating and improving research review processes. Contributing to publications on AI in public health.
-
2020 →
Founder & AI Architect · DENEM Labs
Private LLM infrastructure, compliance reviews and agentic systems for enterprise clients. Delivered 25+ AI projects across LLMs and computer vision, and lead R&D toward safer AI with a small team of scientists.
-
2025-2026
Cofounder & CTO · Loops
Cofounded a venture-backed startup that raised a round including Kima Ventures.
Reference Jules de Villeplée, Cofounder
-
2022-2025
Software Engineer · EDF SEI
Built EDF's Data Office platform on Google Cloud, at Europe's largest electricity producer. Wrote the ML algorithms and prompt-engineered a legal-monitoring agent that tracks new laws affecting EDF (CrewAI, Google Cloud), alongside the data engineering and operational tooling.
Reference Stéphanie Delaunay, Chief Data Officer
-
2021-2022
R&D Consultant · Research to Business
Authored the technical and state-of-the-art sections of R&D tax-credit (CIR) dossiers documenting industrial ML systems, and evaluated deep-learning models across 15+ projects.
Reference Delphine Chatard, Director
-
2021
Founding AI Engineer · Gloth
Built NLP medical-diagnosis models in production (PyTorch, NLTK) for an AI diagnosis startup, and kept the service running (Kubernetes, Django).
Reference Michael Lahmi, CEO
Research
Viability and Performance of a Private LLM Server for SMBs: A Benchmark Analysis of Qwen3-30B on Consumer-Grade Hardware
Reviewers Théo Lepage, PhD · Victor Miara, ex-NVIDIA
Inferring mental states and the humanity we attribute to conversational agents
Reference Olivier Codou, Associate Professor
Talks
- 2026 CREA Geneva · AI Production Innovation: conversational applicationsReference Raphael Troussier, Program Director
- 2025 Canal+ · The future of AI in 2025Reference Mohamed Lemine Beydia, Global Head of AI Factory
- 2025 Van Cleef & Arpels · AI for HR transformationReference Hugues de Pins, Managing Director
- 2025 Cofidis · Landing AI successfully (field feedback)Reference Nicolas Wallaert, CEO
- 2025 AMGEN · Generative-AI workshops for doctors, across FranceReference Emma Lacombe, Services Strategy Lead
- 2022 Podcast IA · #15 · Invited guest on building an AI career independentlyReference Morgan Gautherot, PhD
- 2019 Pass the Salt · Applying social psychology to cybersecurity, presented to a security-expert audience at 19Reference Christophe Brocas, Cofounder
Education
- 2025 First-aid instructor certification (PSC) · Croix-Rouge française
- 2024 Google Cloud Professional Data Engineer
- 2024 IELTS · English proficiency (C1)
- 2022 RNCP level 7, Machine Learning Engineer · CentraleSupélec & OpenClassrooms
- 2021 RNCP level 6, Data Analyst · ENSAI & OpenClassrooms
- 2021 MSc Informatics & Analytics (one semester) · UNC Greensboro
- 2021 Licence, Psychology (Cognitive Neuroscience) · Université de Rouen
- 2020 Data Science: Visualization · HarvardX (edX)
- 2019 Machine Learning: A Case Study Approach · University of Washington (Coursera)