Software Engineer @ Refractal · Computer Science @ UCL

I'm interested in what happens underneath an AI application: how models compute, how inference runs, and how to make the surrounding systems reliable. I work across model internals, AI infrastructure, and agent security.

Open-source contributor to OpenSRE and PostHog.

Model internals — Residual-stream analysis, activation hooks, refusal-vector steering, and mechanistic interpretability.

AI infrastructure — GPU-backed model serving, streaming APIs, evaluation pipelines, observability, SDK integrations, and containerised services.

Agent systems — Stateful execution, tool orchestration, multi-agent coordination, runtime policy enforcement, and adversarial testing.

_{Python · PyTorch · FastAPI · Modal · Docker · SQL · LangGraph · TypeScript · React}

Refractal — Building infrastructure for securing AI agents: runtime controls, evaluation, SDK integrations, and auditable execution.

Hypercast 4D · Research in progress
Exploring hypercomplex neural networks for multivariate time-series forecasting: implementing quaternion and related algebraic layers in PyTorch, and studying their behaviour against real-valued baselines.

Avenir-UX — Researching multimodal agents for simulated human web interaction and automated usability evaluation at UCL Nexus Labs. Pre-print ↗

ReadMyMind — Detects prompt injections in a model's residual stream and steers generation toward refusal.
_{🥇 1st · Cursor Cybersecurity Hackathon}

Sentinel — Decentralised agents negotiate satellite collision avoidance using live orbital data and ML risk classification.
_{🥇 Environment Track · NVIDIA GTC Hack for Impact · Dell GB10 prize}

Sentrix — A multi-agent surveillance and forensics system for detecting and investigating unsafe agent behaviour.
_{🥇 1st · AIEngine UCL BUILD AI Festival · NVIDIA GTC 2026 trip
🏆 Winner · Stanford ASES London · Stanford ASES trip}

Onflow — Multimodal browser agents simulate real users and turn their navigation into UI/UX feedback.
_{🥇 1st · Bloomsbury Startup Academy Demo Day
🥈 2nd · Gemini 3 London Hackathon}

_{Exploring mechanistic interpretability, reinforcement learning, world models, and alignment. Always up for a conversation about the systems behind AI.}