- Building and maintaining a unified WordPress theme and reusable UI components in React and TypeScript, standardising front-end design across 1,200+ web services and experiment sites at CERN.
- Developed a Python migration tool that transitioned 1,000+ sites and 100,000+ pages from Drupal to WordPress, cutting per-site time from hours to under one minute.
- Developed and Deployed NeutrinoReview, a full-stack AI platform (Python & React) automating systematic literature reviews for medical research in collaboration with WHO.
- Implemented and conducted research on LLM-based calibration and screening approach to improve classification accuracy across research corpora.
- Led a team of 3 developers to develop Arbiter, a SaaS social listening platform used by civic organisations including Nest fact-checkers (Mongolia) and DW Germany.
- Engineered scalable data pipelines processing 1B+ data points; built semantic search using PostgreSQL and pgvector.
- Developed NLP pipelines and an AI assistant for automated trend analysis; deployed on AWS and GCP.
GPT language model from scratch. Multi-head self-attention, positional encoding, character-level tokenisation. 0.5M parameters, trained on Shakespeare.
Autograd engine with backpropagation over dynamic computational graphs. Trained an MLP for binary classification from first principles — no ML libraries.
Full LC-3 architecture in C: 16-bit instruction set, trap routines, fetch-execute cycle with computed GOTO. Runs real assembly binaries.
AI platform automating systematic literature reviews at CERN. Integrates ArXiv and PubMed; LLM-based screening with calibration pipeline.
Reverse-engineered the Moj social media API via encrypted traffic interception. Built a concurrent analytics pipeline and Flask dashboard without official access.
Full-stack social platform for college students: auth, posts, comments, real-time feed. First project — the Zuckerberg moment that didn’t quite happen.