I’m a Machine Learning Research Engineer with a background in Applied AI, Computer Science, and Genetics.
I taught myself to program at age 12 so I could make robots and haven’t stopped building since. Starting in 2018, I’ve worked on ML Engineering and AI Research across a variety of domains and organisations, from startups to governments.
My hobbies include indie filmmaking, music production, technology, and sci-fi. I can usually be found combining a few of these at a time.
Check out my projects, essays, and book reviews. You can subscribe by RSS or follow my updates at the links above. If you’re interested in any of my projects, want to work with me, or just think we’d get along, please do reach out on Twitter / X, Bluesky, or by email.
Some background:
- At the end of 2024, I completed a 12-week batch at the Recurse Center, honing my craft and making fun things. I focused on performance browser graphics, on-device LLMs, learning Rust, and building command-line tools.
- In 2022, I co-founded a startup called Bountyful AI, backed by Europe’s leading accelerator, EF.
- As Lead ML Engineer, I developed our MLOps platform for Generative AI: Enabling companies to observe, own, and optimise LLMs via human feedback and knowledge distillation.
- My work enabled early GPT-3 and 4 adopters to optimise for cost, latency, and performance by distilling to open models on our scalable infrastructure.
- Demonstrated up to 8x cost reductions and 3x speedup on LLMs via distillation, supervised finetuning, and RLHF.
- Previously, I worked in a variety of machine learning roles. I’ve developed, trained, evaluated, and deployed ML systems at terabyte scale across scientific, biomedical, financial, and industrial applications.
- I also used to moonlight as a self-taught quant trader.
- My academic research has involved applying AI to biomedical sensors, synthetic data, and generative models.
- I developed TableDiffusion, the first diffusion model for privatised tabular data.
- My MSc thesis examined methods for synthesising privatised (healthcare) data using generative models like GANs, autoencoders, and denoising diffusion models.
- I’ve applied ML to Quantified-Self projects and used Genetic Programming to evolve automated dosing models for the anticoagulant drug Warfarin.
- I received my MSc in Artificial Intelligence from VU Amsterdam x University of Amsterdam, graduating Cum Laude. I also hold a BSc in Computer Science (Honours) and Genetics, graduating First Class and tied top of the year.
- I was co-creator and co-host of the Bit of a Tangent podcast with my long-time friend Jared, where we had technical conversations about AI, Neuroscience, Mental Models and how to live better lives. We published 31 episodes with 30,000+ downloads.
If you’re interested in any of my projects, want to work with me, or just think we’d get along, please do reach out on Twitter / X, Bluesky, or by email.
Projects
Highlights
TableDiffusion
A deep learning algorithm I developed for training diffusion models on tabular data under differential privacy guarantees.
Quantified Sleep
This project applied statistical learning techniques to an observational Quantified-Self (QS) study to build a descriptive model of sleep quality. A total of 472 days of my sleep data was collected with an Oura ring. This was combined with a variety of lifestyle, environmental, and psychological data, harvested from multiple sensors and manual logs.
Watch and learn
Winning EF’s Bio x AI hackathon with a multimodal LLM for lab protocol automation.
FOMO: LLM embeddings to keep up with AI
An open source newsreader app I built using LLM embeddings to semantically prioritise arXiv and Github updates.
AImong Us: a reverse Turing Test game
Can you outwit a gang of LLMs?
Recent
Notes for learning JAX
As a learning exercise, I recently implemented simple-jax-nn, a simple neural net for the MNIST dataset, written with JAX. This post is my notes on using JAX (as a PyTorch / NumPy user) from working through the JAX docs and a NN tutorial. Simple JAX neural net I managed to get it to use my MacBook’s GPU/Accelerator through the Metal support in jax-metal. It’s a little bit tricky, but will prove useful in future transformer projects that can make use of the hardware. ...
Reflections on Vibecoding
A sorcerer’s guide to wielding chaotic and powerful magicks.
Music Melee: a high-speed parkour FPS for making beautiful sounds
Embracing the exponentials and vibecoding an entire 3D game in a few days with Aider, o3-mini, and Claude 3.7 sonnet.
DevLog: orbital mechanics game in your browser
The development stories and roadmap for Orbital, my physics-based 3D simulator for realistic orbital maneuvers that runs in your browser with WebGL and Three.js.
Salvador DALL-E: Falling for Svelte and serverless shennanigans
How I learned Svelte and outsmarted Vercel to make a delightful frontend to OpenAI’s DALL-E … just for my mom.
Streams of beautiful riches
Enter richify.py: a real-time Markdown rendering tool that supports streaming input. Built with Rich and shipped effortlessly with uv.
What I made at the Recurse Center
Links to the projects, games, blog posts, and tools I shipped in my 12 weeks at RC.
DevLog: llmpossible
How I built a command-line LLM on Apple Silicon for RC’s ‘Impossible Stuff Day’
Raytracing from scratch in pure JS
A dynamic realtime raytracer in pure JavaScript you can run in your browser.
Building PorePatrol LLM at the Anthropic Hackathon
We used LLaVA and Claude to build a crazy multimodal LLM health assistant. OpenAI made it redundant the next week.
Learning quant. trading with ML on Numerai
Making 95% annual return on the hardest data science tournament in the world, for fun and profit.
Generative models for synthesising private datasets
My MSc thesis and what it aims to do for biomedical science.
Reviewing my 2020 predictions
How I almost outsmarted 3 pro forecasters using Bayes rule.
Gianluca's Predictions for 2020
Trying to out-forecast Vox in 2020. What could go wrong?
My older projects
A selection of my older projects (pre ~2019) transferred from an old website for posterity.