Simon Coessens
I'm a 24 year old developer and AI researcher living
in Paris, France.
I recently graduated with Highest Honours from CentraleSupélec (Université Paris-Saclay) as part of the Erasmus Mundus BDMA program. My thesis received the Master Thesis – Industrial Impact Award.
I also hold degrees in Computer Science and Mathematics from KU Leuven. My research focuses on 3D scene & world generation, multi-view diffusion, 3D Gaussian Splatting, NeRFs, and novel view synthesis, alongside applied work on agentic AI systems.
Recent projects
What I've been working on
MultiViewPano: Training‑Free Panorama Generation
Spotlight talk at ICCV 2025 (SP4V workshop). Training‑free 360° panorama generation from arbitrarily posed inputs using SEVA multi‑view diffusion and a pose‑aware stitcher. Ongoing research.
Agentic Video Generation Pipeline (Amadeus)
Developed an agentic pipeline for video generation during my internship at Amadeus, using LangGraph and experimenting with available video generation models to orchestrate automated images‑to‑video workflows.
DSA compliance tool (Agentic AI)
Prototype agentic copilot for DSA compliance, built with Prof. Antonio Davola and Vittorio Franceze to blend legal expertise with LangChain + GraphRAG automation. On track for a broader launch, with the next milestone being the DSA & Platform Regulation Conference 2026 (University of Amsterdam).
GaussianDiffusion: 2D Gaussian Splatting Latent Diffusion
Diffusion-based image generation operating in a latent space of 2D Gaussian splats. Research project @ CentraleSupélec; presented at JDSE 2025.
Other projects
Here are some key projects I've been involved in:
Revolut Revpoints-Plus 🔗 - Compact proof-of-concept that makes Revolut Points a daily habit by recommending vendors where paying with points is a no-brainer. Won 3rd place at the UPC x Revolut hackathon.
ScholarIA 🔗 - An all-in-one research assistant for streamlining academic work with paper management, personalized recommendations, and note-taking.
Neurocrypt 🔗 - Privacy Preserving AI Hackathon project on secure EEG processing with Fully Homomorphic Encryption for real-time analysis.
A.N.N.A. 🔗 - An AI assistant designed to help developers and organizations assess compliance with the EU AI Act. It won the 1st Prize and the Most Innovative Solution award at the LEADS Innovation Challenge. The full codebase is open-source on GitHub for transparency and reuse.
Some pictures
