Description

An AI developer of more than 10 years, I’ve built: RL algorithms for the control of autonomous vehicles (Oxbotica, an Oxford University spin off company), power stations (Centre for AI, SSE, a big six energy company); Forecasting models for wind farm generation, energy demand, energy prices (SSE, Shell); Computer vision algorithms for the maintenance of energy assets (distribution networks, drilling platforms) (SSE, Shell); Multimodal Language-Vision models for generating insurance claims from images and videos of car accidents (Core Research Team, Tractable AI, unicorn deep learning startup); RL and VLAs for robotic control and VLMs for robot evaluation (Leap AI, robots packing produce for large supermarkets).

Languages

English
Native or bilingual

Workplace preferences

Can work on-site

London (up to 50km)

Blend360
AI Engineer
June 2025 - September 2025 (3 months)
Solutions
• I developed an agent to extract structured transaction information from free form descriptions and cross reference with a master table using vector search.
• Scaled the solution using Spark map-reduce to optimally process large data volumes (millions of records).
• Developed to be production ready for imminent Diageo deployment.
Gen AI
Leap Automation,
Lead AI Scientist
January 2024 - May 2025 (1 year and 4 months)
Solutions
• I developed and deployed imitation learning, reinforcement learning and vision language action (VLA) algorithms to control robot arms packing fresh produce at very high, industrially relevant, accuracy and picks per minute.
• I built, trained and deployed vision language model (VLM) and active learning based monitoring and decision systems to autonomously evaluate and manage robot performance.
• Developed generative AI vision models, based on diffusion, to create large artificial datasets to enhance the performance of supervised models.
• All algorithms were deployed using MLOps and CI/CD
• Oversaw projects, including Innovate UK funded future of robots research.
Shell,
AI Scientist
November 2022 - January 2024 (1 year and 2 months)
Solutions
• For the AI dept I developed and productionised an agentic LLM pipeline to enable querying of Shell internal documents, databases and internal trading tools with natural language. This included a dynamic knowledge graph memory, retrieval augmented generation (RAG) and tool use. The full pipeline was optimised via RL.
• This saved traders and researchers on average 2 days a week in information compilation time.
• For trading I developed and productionised an agentic AI that used RL to train an LLM to reason about news and future events (significantly outperforming humans).
• I also developed and productionised price forecasting models based on deep learning.
• The systems developed were used across the trading floor by 100s of traders and trading managers.
• All algorithms were deployed using MLOps and CI/CD