Selected work

Projects

Projects are listed with their current state and source code. GitHub contains the full history, experiments, and smaller repositories.

ML project series

A numbered series of focused machine learning projects. Projects are added here as the work is completed.

01

Transformer from scratch

Complete

A character-level GPT-style transformer implemented in PyTorch, including attention, positional encoding, training, evaluation, and generation.

Source ↗
02

Data analysis and scientific method

Complete

AG News analysis with pre-model hypotheses, TF-IDF and logistic regression, error analysis, ROC/AUC, and LIME. Test accuracy: 91.78%.

Source ↗
03

News classifier API

Complete

The Project 2 model served through FastAPI with Pydantic validation, tests, Docker, model documentation, and a Railway deployment.

Source ↗
04

Next project

In progress

The next entry will be added when there is working code and something useful to report.

Other work

Research, local tools, learning environments, and hardware experiments.

Tactical Style Fingerprint

Five transparent event-data dimensions for comparing Premier League teams, with tactical neighbours and grounded AI explanations.

Source ↗

Hinglish hate-speech research

Cross-dataset comparison of mBERT and MuRIL for code-mixed language.

Source ↗

Lumen

A local-first macOS desktop agent with local model discovery and safety confirmations.

Source ↗

TwistLab

An interactive WebGL environment for learning twisty-puzzle notation and algorithms.

Source ↗

Drishya Physics

An interactive mechanics lab with an analytical solver and conservation tests.

Source ↗

Football Rule Cite

A retrieval tool that explains football decisions using the IFAB Laws of the Game.

Source ↗

ESP32 touch diagnostic

A raw coordinate and pressure diagnostic for an ESP32 resistive touchscreen.

Source ↗

EngineerCheck

A recent TypeScript project exploring practical engineering evaluation.

Source ↗

More