Australian data scientist building AI agents, solving hard problems and relentlessly learning.
I like a clear target and figuring out how to get there — through modelling, optimisation, and lots of testing and iteration.
Find me on X · Join the Discord · Watch the Minecraft run
I built mine-ai-mcp, an MCP server that lets an AI model request actions in Minecraft and get useful feedback on what happened.
The agent completed vanilla Minecraft Java 1.21.4, Survival, Normal difficulty in 3h 56m, with one death and a successful gear recovery.
The engineering underneath:
- A TypeScript navigation system inspired by Baritone
- A combat state machine for fast reactions, steered by model policy
- Fixes to underlying Mineflayer libraries
- Useful feedback and errors so the model can make informed decisions
- Lots of testing, fixing and iterating
Explore the code · Watch the full run · Run artifacts & world save
| Area | What I work with |
|---|---|
| Data science & modelling | R, Python, gradient-boosted models |
| Data processing | Spark, Polars |
| Software & AI agents | TypeScript, MCP, agent tools and feedback systems |
| Problems I enjoy | Optimisation, navigation, and making systems more reliable |
| Currently learning | Rust |
Building, experimenting, and sharing what I learn along the way.
