A collection of projects I’m working on or have completed - this is a bit of an info dump, this page needs work!
In Progress
BluffSport (name is still being worked on!)
A conversational assistant for the sports illiterate; inspired by “BluffBall” of the IT Crowd TV show. I’ve been meaning to build this for years, and it’s finally taking shape. Soon I’ll be able to talk my way through any pub conversation about sport.
This is a bit of a flagship project, I’m building this as a collection of Rust-based services, focusing on:
- Architectural design and infrastructure
- Service-oriented architecture
- Event-driven design
- Kubernetes
- Fully automated deployment (Flux, Terraform/Tofu, etc.)
- Data gathering
- Web crawling recent sports events and news
- Processing and storing for both training and retrieval-augmented generation (RAG)
- Text generation
- Utilise an LLM combined with RAG to create conversational topics on recent sports events
- Web front end
- Using SvelteKit
- API to back end infrastucture, “Backend-for-frontend”
- Automated testing with Playwright
There’s plenty of scope for expansion once the foundation is solid, but for now I’m prioritising stable infrastructure and basic text generation from the gathered data.
Complete
Neuronet - A Neural Network from Scratch
Building a neural network library from the ground up in Rust to better understand the fundamentals. No frameworks, no shortcuts - just gradients, backpropagation, and a chunk of matrix multiplication.
This was quite a fun excercise and I cover writing and testing it in my blog; in pracice though it doesn’t make sense to write this myself as there are some great industry standard tools for this.
[github.com/lordmoocow/neuronet]
RAG - A naive approach
I initially created this as part of a post on basic a RAG pipeline that I intended to use as a starting point for another project. I used Claude Code to create a basic template as a tutorial of the implementation, and hand coded the details.
It worked - but it isn’t great. I’ve since leaned into the ever increasing power of Claude Code to create a much more comprehensive chunking library (post on the coming soon).
[github.com/lordmoocow/ragsport]
Planned
Flocking Boids
To be honest, this is something I’ve implemented so many times in different languages and architectural designs over the years. It’s my “go to” project for trying out a new language in anger, fun visuals of emergent behaviour.
This time I’d like to try exploring whether flocking simulation data could be useful for training models to predict emergent collective behaviour. Still very much in the “interesting idea” phase but definitely something that keeps my brain whirring away at night.
This is always subject to change!