Hi, I'm Marc Monfort.
Doctoral researcher at Aalto University, working on multi-agent reinforcement learning.
My PhD is on social intrinsic motivation: building artificial agents that are more reliable, proactive and trustworthy. Before that I spent four years at the Barcelona Supercomputing Center, building dataClay, a distributed object store for HPC.
Below are my research and side projects, from federated learning on microcontrollers to game engines built from scratch.
Projects
I built a lightweight mixed reality prototype in Shanghai to overlay historical context and urban design ideas on Fuxing Island, as part of Tongji University’s international summer school.
A step-by-step log of how I designed, built and self-hosted my personal portfolio—covering everything from choosing Astro to the domain, dynamic DNS, Proxmox virtualisation, Nginx Proxy Manager and automated HTTPS.
dataClay is a distributed data management middleware designed to enable object-oriented storage and computation across multiple nodes for high-performance and cloud environments.
On-device training of a keyword-spotting model on Arduino Nano 33 BLE Sense boards, extended with federated learning. Covers the prototype, the experiments and results from our Electronics and ACM GoodIT papers, and the open-source code.
Hands-on experience running deep learning workloads on the MareNostrum5 supercomputer at BSC. Covers optimizer benchmarks, MNIST improvements, multi-GPU scaling, and how HPC changes the workflow compared to local environments.
Technical exploration of cooperative multi-agent deep reinforcement learning with MATD3, tested in PettingZoo's Simple Spread and Simple Speaker Listener environments. Includes algorithm pseudocode, actor-critic diagrams, training GIFs, and experimental results.
Implementation and experiments with PPO agents on Pendulum, BipedalWalker, and LunarLanderContinuous environments, including training results, episode gifs, and algorithm insights.
Technical demonstration of Comp, a custom game engine project showcasing real-time rendering and basic physics built from scratch.
Technical demonstration of Break-In, a first-person stealth game prototype featuring lockpicking, custom physics, and rendering engine built from scratch.
Multi-agent system for smart city traffic simulation, integrating vehicles, traffic lights, and cloud agents with semantic reasoning and negotiation protocols.
A Haskell implementation of Connect Four, featuring multiple AI strategies—including minimax with heuristics—for an interactive terminal-based game.
Doctoral Researcher