Machine learning engineer and data scientist specialising in ML for language and vision, including large language models.
I work for Netcall as a founding member of the AI team, building ML solutions for the company’s Liberty suite with a focus on LLMs, evaluation, RAG, and agentic systems. Prior I worked for Kainos, delivering language and vision AI/ML projects for a range of clients including in defence and insurance.
On top of my job I work on open-source AI projects and research. Some of the most interesting are:
- Currently, Reasoning Gym. We are building procedural dataset generators of algorithmically verifiable problems, for training language models to reason via reinforcement learning techniques such as GRPO. You can read a preprint of our paper on arXiv.
- Previously, Open Assistant. We built the first open-source alternative to ChatGPT, published a dataset of high-quality instruction data, and trained LLMs using supervised finetuning (SFT) and reinforcement learning from human feedback (RLHF). Our paper was accepted to NeurIPS 2023.
Outside of work I take interests in economics, policy, and football.
If you’d like to get in touch, feel free to reach out via LinkedIn. You can also view my Google Scholar profile.
Latest Posts
-
Reasoning Gym Paper
Evaluation and RL training experiments with Reasoning Gym -
Intro to Reasoning Gym
Building procedural data generators to train reasoning models -
First Open Assistant Models and Dataset Release
New open-source language models tuned to follow instructions
Subscribe via RSS