bet365
Mar 2024 – present
Manchester (hybrid)
- Applied AI Engineer, AgentOpsApr 2026 – present
- Software Developer, InnovationApr 2025 – Mar 2026
- Junior Software Developer, InnovationMar 2024 – Mar 2025
I find slow processes with teams across the business, agree the KPIs, then build, ship and support the agent that fixes them.
- Built a fraud investigation agent used by around 250 people in fraud teams across the UK and Malta. It analyses a flagged user's site activity, highlights high-risk behaviour and gathers the evidence for the investigator's report. Time per case fell from 3–4 hours to about 30 minutes, and investigators keep the final decision.
- Delivered an AI translation workflow for the international content team. Drafts are generated using the team's own term dictionary, then reviewed and corrected by translators. Localisation finishes about 4 months sooner, with fewer external contractors.
- Currently on game releases, a company-wide effort to cut a process that takes over a month per game. Building a document repository and tracking system with AI certificate scanning and game scoring.
- Design and write the agents myself, mostly as plain Python, TypeScript and C# with some Google ADK, using RAG and tools from bet365's in-house MCP platform.
- Run evals on each workflow to compare GPT, Claude and Gemini models on output quality and price before picking one.
- Ship each agent as a Docker image through Artifact Registry to GKE, with config as code in Anthos Config Management.
- Host and maintain bet365's n8n instance, used by around 25 teams. Onboard teams and review their workflows before they go to production.
- Spent my first eight months on commercial threats, building anti-scraping protections for bet365's odds data, then moved to AI work and built about 10 proofs of concept for 5 business teams.