ML researcher building clinical and scientific AI tools
Reinforcement learning, medical AI, scientific computing, and LLM tooling
Hi, I'm Thomas (you'll find me as CTCycle around here). I'm an ML researcher and data scientist building applied AI tools for clinical and scientific workflows. My background is a bit of a winding road: started in biotechnology, picked up a PhD in engineering sciences, and somewhere along the way traded the lab bench for a terminal. Best decision I ever made.
A quick version of the story: bachelor's and master's in Biotechnology from the University of Genoa, then a PhD in Engineering Sciences at the Vrije Universiteit Brussel, where I worked on the CheckPack project, developing micro-sensors to detect food spoilage and monitor quality in real time. That project taught me two things: how to design and run real-world experiments, and that I really, really enjoy turning data into something useful. After the PhD, I doubled down on machine learning, and I have been building ML tools ever since.
These days, my work spans reinforcement learning, medical computer vision, LLM-powered clinical copilots, computational chemistry, and whatever interesting intersection of biology and computation comes up next. Everything on this profile is open source, because science should be reproducible and good tools should be accessible.
Browse all 18 projects in the HTML Catalog — a searchable, filterable directory with stats, tech tags, and direct links to every repo. If you prefer your projects as floating islands in 3D space, there is a 3D viewer too (yes, each project gets its own island shape).
Reinforcement learning (DQN) agent that learns to play roulette. Full-stack application: PyTorch training pipeline + FastAPI backend + React frontend + Tauri desktop shell.
Reinforcement Learning PyTorch React Tauri
Vision-language model that generates descriptive radiology reports from chest X-ray scans. Built on transformer architectures, trained on clinical datasets.
Computer Vision NLP Transformers Clinical AI
ML-driven adsorption isotherm fitting and prediction using NIST/ARPA-E databases. Fits theoretical models to empirical data and predicts gas uptake for materials science applications.
SciML Chemical Engineering NIST Databases
| Project | Area | Stars | Tech |
|---|---|---|---|
| FAIRS Roulette Player | RL | 6 | PyTorch, React, Tauri |
| XREPORT Radiological Reports | Medical AI | 6 | Python, Transformers |
| EMADB Autopilot | Clinical Data | 6 | Browser automation |
| ADSMOD Adsorption Modeling | SciML | 3 | Python, RDKit |
| FEXT Autoencoder | Computer Vision | 1 | PyTorch |
| DILIGENT Clinical Copilot | Clinical LLM | 0 | LLM, FastAPI |
| TKBEN Tokenizer Benchmarker | LLM Tools | 0 | Python |
| ParaGraph LLM Workflow | LLM Tools | 0 | React Flow |
| AEGIS Geospatial View | Geospatial | 0 | Python, Web |
| Domain | Technologies |
|---|---|
| Languages | Python, Java, JavaScript, SQL, Bash |
| ML / DL | PyTorch, TensorFlow, Transformers, HuggingFace, LangChain |
| Infrastructure | Docker, Kubernetes, Linux, GitHub Actions, REST APIs, PostgreSQL |
| Scientific | RDKit, NumPy/SciPy, Pandas, NIST databases, ARPA-E |
| Tools | Git, VS Code, IntelliJ, Postman, Jupyter, CI/CD |
- Building: Open source ML tools at the intersection of clinical research and artificial intelligence. RL agents, medical image models, LLM copilots, computational chemistry pipelines.
- Learning: Front-end development (UI/UX is harder than it looks) and productionising ML with Rust.
- Looking for: Interesting collaborations at the crossroads of biotech, healthcare, and machine learning. If you have a project that mixes biology and code, I would love to hear about it.
- Ask me about: Reinforcement learning, medical AI, turning a PhD in biotech into an ML career, or how to get started building things that actually ship.
