Skip to content
View mathisdelsart's full-sized avatar
1 follower · 0 following

Achievements

Achievement: QuickdrawAchievement: Pair ExtraordinaireAchievement: Pull Sharkx3

Achievements

Achievement: QuickdrawAchievement: Pair ExtraordinaireAchievement: Pull Sharkx3

Block or report mathisdelsart

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
mathisdelsart/README.md

Mathis Delsart — AI / ML Engineer

LinkedIn Email Open to Opportunities

Typing SVG

Wave About Me

I recently graduated Summa Cum Laude from UCLouvain (École Polytechnique de Louvain) with an MSc in Computer Science & Engineering, specializing in Artificial Intelligence.

I'm looking to join a team as an AI / ML Engineer, building AI systems that actually ship to production.

I'm passionate about Computer Vision, Natural Language Processing, and Generative AI (LLMs, RAG & Agentic AI) — and I enjoy engineering the entire pipeline behind them, from large-scale training on HPC GPU clusters (SLURM) to Docker-containerized deployment on Azure and full-stack delivery.

I'm open to AI / ML Engineer roles in Belgium (on-site/hybrid) or fully remote across the EU.

View My CV

Reach out via LinkedIn or at mathis.delsart@gmail.com.

Tools Tech Stack

ML & DL

Generative AI

Web & Backend

Databases

DevOps & Cloud

Systems & Embedded

Dev Tools

AI Tooling

Research Research — Master's Thesis

Deep Reinforcement Learning for Competitive Agents in MicroRTS — Architecture, Training, and Tournament Evaluation

UCLouvain · Completed June 2026  ·  Accepted at IEEE CoG 2026

A DRL agent for real-time strategy games, fusing a U-Net spatial encoder with an entity-level Transformer (UECD) and trained end-to-end with a modular PPO pipeline on HPC GPU clusters (SLURM) — every design decision individually ablated.

  • 🥇 96.67% pool win rate — tops a 19-agent tournament (1st on 4 of 5 metrics)
  • ⚔️ Beats RAISocketAI (CoG competition winner) in 65.7% of head-to-heads
  • ⚡ Trained on just 9.47 GPU-days — vs the 23.6 reported by RAISocketAI

Live Site Paper Dissertation

Deep Reinforcement Learning Proximal Policy Optimization Self-Play Behavior Cloning
Reward Shaping Transformer Game-Theoretic Evaluation MicroRTS Real-Time Strategy

UECD-Best vs RAISocketAI

UECD-Best (ours, top-left) vs RAISocketAI (bottom-right)

📂 View the repository →

Projects Featured Projects

Deep Learning & Computer Vision
Project Repo Description Stack
Satellite-Image-Segmentation View Satellite image segmentation with supervised & self-supervised learning Python PyTorch NumPy
3D-Hand-Gesture-Classification View Comparative study: KNN, LSTM, DenseNet, HMM for 3D gesture recognition Python TensorFlow scikit-learn Pandas
GANs-Diffusion-Kernels View Advanced generative models: GANs, diffusion models, kernel methods Python PyTorch Diffusers Transformers Jupyter
Heart-Failure-Detection-Smurf View Heart-failure risk prediction on a fictional "Smurf Society" dataset (synthetic, not real clinical data) Python PyTorch scikit-learn Pandas Jupyter
ML-Classification View Supervised learning models for medical & biological datasets Python scikit-learn Pandas Jupyter
Celebrity-Dataset-Explorer View Interactive face attribute analysis with clustering & dimensionality reduction Python Dash Plotly scikit-learn Pandas
Natural Language Processing & LLMs
Project Repo Description Stack
Sourcio — RAG Study Assistant View Deployed RAG tutor answering only from your own courses — citation-by-construction (the model never sees a page number) + a three-layer refusal guard. Live app: 100% citations, 94% refusal accuracy Python LangChain FastAPI Qdrant Next.js Docker
NLP-Models View Progressive NLP pipeline: Question Classification (Naive Bayes), Vector Semantics (TF-IDF/PPMI/Word2Vec), BERT-based QA (SQuAD) Python NLTK Gensim PyTorch Transformers
Text-Prediction-TwitOZ View Intelligent text prediction system using N-grams algorithm with real-time word suggestions Oz/Mozart N-grams
Game AI & Reinforcement Learning
Project Repo Description Stack
Snakes-Ladders-MDP View Optimal strategy via Markov Decision Processes & Q-Learning Python NumPy Matplotlib
AI-Odyssey View Search algorithms: uninformed to adversarial game-playing agents Python Pygame PyCSP3 MiniSat
AI-Connect4-Game View Connect 4 with Minimax AI — Can you beat a machine that thinks 4 moves ahead? Python Pygame NumPy
Data Science & Analytics
Project Repo Description Stack
Data-Science-Portfolio View Applied statistics & data science: energy analysis, star classification, species clustering Python Pandas scikit-learn SciPy Jupyter
Pattern-Mining View Pattern mining: frequent itemsets, sequential patterns, and Bayesian networks Python NumPy Pandas scikit-learn
Database-SQL View SQL query optimization and Neo4j graph analysis with Cypher SQLite Neo4j SQL Cypher
Systems Programming & Compilers
Project Repo Description Stack
Custom-Compiler View Full compiler: custom language → JVM bytecode Java Gradle ASM JUnit
Prolog-Python-Analyzer View Prolog-powered static analyzer for Python code detecting 13+ code smells through declarative pattern matching on ASTs Prolog Python SWI-Prolog
Scheme-OOP View Building OOP from scratch in pure Scheme: classes, inheritance, and polymorphism from closures and message passing Scheme Racket R5RS
Java-COP View Context-Oriented Programming in Java using reflection, dynamic proxies, and bytecode instrumentation with custom profiler agent Java Javassist Python
Low-Level Systems & Embedded
Project Repo Description Stack
CORDIC-FPGA-Accelerator View High-performance FPGA accelerator integrated with ARM processor SystemVerilog Python ARM Quartus
High-Performance-Server View High-performance HTTP matrix multiplication server on NGINX with progressive optimizations: cache-aware algorithms, SIMD (AVX-512), and CUDA GPU acceleration C NGINX AVX-512 CUDA
Multithreaded-Bellman-Ford View Parallel single-source shortest-path (Bellman-Ford) in C, parallelized with a POSIX-thread pool — ~6× faster on large graphs C POSIX Python
Dynamic-Memory-Allocator View Custom heap allocator with next-fit & coalescing C Make
Concurrent-Bench-Sync-Primitives View Performance benchmarking framework for custom synchronization primitives (TS, TTS, BTTS) vs POSIX on Reader-Writer, Dining Philosophers, Producer-Consumer C POSIX Python Matplotlib
Tar-Archive-Parser View Lightweight TAR archive parser in C — validation, file operations, symlink resolution C POSIX Make
Numerical Computing & Optimization
Project Repo Description Stack
FEM-Bridge-Simulator View Finite element simulator for bridge structures under static loads, featuring 2D elasticity modeling and animations C Python OpenGL CMake Gmsh
Optimization-Algorithm View Solving NP-hard problems: exact methods, relaxations, heuristics Java Maven JUnit
Algorithm-and-Data-Structures View Clean and well-tested core data structures and algorithms with performance focus Java Maven JUnit
Web & Mobile Development
Project Repo Description Stack
Shopping-Cart-Application View Scalable microservices e-commerce platform with custom recommendations, smart ranking, and dynamic shopping Node.js Docker SvelteKit CouchDB Redis Azure
Private-Tutoring View Bilingual, responsive tutoring website, statically deployed to GitHub Pages Next.js React TypeScript Tailwind CSS
Energy-Tracker-MeterBuddy View Android energy-consumption tracker with charts, Room storage, and Firebase sync Kotlin Jetpack Compose Android Room Firebase
Games
Project Repo Description Stack
Chess-Game View Full chess rules, multiple themes, and a clean Pygame interface Python Pygame
Flappy-Bird-Game View Classic Flappy Bird recreation with physics and animations Python Pygame
Modern-Snake-Game View Modern, polished Snake game with gradients and animations Python Pygame
Coursework & Foundations
Project Repo Description Stack
EPL-Chronicles View Aggregated UCLouvain coursework: systems, ML, data science & programming paradigms C Assembly Python Prolog Scheme

Fire GitHub Stats

GitHub Stats

Profile Summary

Pinned Loading

  1. fem-bridge-simulator fem-bridge-simulator Public

    Finite element simulator for bridge structures under static loads, featuring 2D elasticity modeling and animations.

    C 4

  2. cordic-fpga-accelerator cordic-fpga-accelerator Public

    High-performance FPGA-based CORDIC accelerator integrated with an ARM processor.

    Python

  3. shopping-cart-application shopping-cart-application Public

    ElastiCart — a scalable microservices-based e-commerce platform featuring custom recommendations, smart ranking systems, and dynamic shopping experiences.

    JavaScript

  4. microrts-drl-uecd microrts-drl-uecd Public

    DRL agent for MicroRTS: U-Net + entity-Transformer (UECD) policy trained with modular PPO. Tops a 19-agent IEEE-CoG-style tournament at 96.67% WR and beats RAISocketAI in 65.7% of head-to-heads, on…

    Python 1

  5. nlp-models nlp-models Public

    End-to-end NLP projects from probabilistic models to transformers. Covers text classification, word embeddings, and BERT-based question answering. Built with Python, NLTK, Gensim, PyTorch, and Hugg…

    Jupyter Notebook

  6. sourcio sourcio Public

    Sourcio — an AI tutor that answers only from your own courses: always cited, never invented.

    Python