TEORIAT Chess Engine
PyTorchA chess AI trained only on my own Chess.com history. It encodes each move as a compact tuple and predicts my next move, imitating how I actually play rather than how an engine would.
The big builds first, then the small sharp tools, then the ventures and the research.
the builds I most like explaining
A chess AI trained only on my own Chess.com history. It encodes each move as a compact tuple and predicts my next move, imitating how I actually play rather than how an engine would.
Real-time messaging: private and group rooms, read receipts, friend requests, GDPR data controls. Django Channels over WebSockets with Redis pub/sub, deployed on Railway with PostgreSQL.
focused tools and experiments
A scheduled Airflow pipeline that extracts, cleans and validates weather data, engineers daily and monthly aggregations plus wind categories, and flags outliers by threshold before loading to SQLite.
An endless runner you control with your body. Pose estimation reads the webcam and turns lean, jump and duck movements into game input in real time.
A neural net predicting term-deposit sign-ups: 89.3% accuracy and 0.76 AUC over 41,188 records, with batch normalisation and dropout to handle heavy class imbalance.
A small local Mac tool that turns messy copied text into clean, structured, reusable data. Narrow scope, does one thing.
ventures and research
An AI Finnish speaking-practice platform, built through Boost Turku's Startup Journey. Generalist founder role: I built the marketing site and the app prototype.
Machine-learning analysis of used automotive spare-part pricing: what drives price, and estimating market value from raw listing data. Final-year thesis, in review.
Draft copy, written by the assistant. Ritesh replaces it in his own voice.