Software

Current projects

corpusforge

Data-optimisation toolkit: distributed ingestion, deduplication, quality filtering, contamination detection, and matched-compute fine-tuning evaluation. In development.

fse-ragbench

Retrieval and generation evaluation harness, with a benchmark for retrieval-augmented generation over public fire-safety guidance. In development.

git4kolibrios

A Git client for KolibriOS, an operating system written largely in x86 assembly. The target is clone over HTTP, written in C against the KolibriOS SDK, which means SHA-1, zlib inflation, packfile parsing and delta resolution all have to be built from scratch. In development.

Earlier projects

Real-time fire and smoke detection from video

MSc dissertation project. Transfer learning with MobileNetV2 and MobileNetV3-Large (TensorFlow/Keras, OpenCV) on a 651-image dataset derived from MIVIA fire-detection video sequences. A PyQt5 operator console routed model alarms to a human reviewer before escalation. The dataset is small, and the size is stated because it bears on what the results can support.

PrerokGlobal

Team project. Shipment and logistics platform. Team of six; I led backend and DevOps. TypeScript, Node.js, Express, MongoDB, Docker and CI/CD.