Economics & Computer Science at Boston University, now an MS in Business Analytics at BU Questrom. I build things where the data has to turn into a decision — pricing, forecasting, retrieval, reinforcement learning — and I'm looking for software, data and AI roles starting 2027.
🌐 otawfik.com · 💼 LinkedIn · ✉️ otawfik@bu.edu
| Project | What it is | Result |
|---|---|---|
| Risk-Deep-Q-Learning | Dual-decoder deep Q-network that plays the board game Risk (Java) | 9/9 vs. the heuristic agent; 72% head-to-head |
| FootSnap | A football card game on live data: ~3,000 player cards, a turn-based duel vs. an AI, Poisson match forecasts and a 3,000-run season simulator (TypeScript, React Native) — open it | Walk-forward backtested forecasts; 774 licensed photo cut-outs |
| Cosmic Explorer | An 8-planet arcade game where every frame is explicit linear algebra — play it | No game engine |
| Fridge Vision | YOLOv8 finds what's in your fridge, a trained classifier scores freshness, TF-IDF suggests recipes | 97.5% freshness accuracy |
| PricePilot | Per-SKU demand elasticity via log-log regression, profit-optimal prices via the Lerner index | +24.6% profit on a simulated panel |
| Forecast Studio | Six forecasting models, walk-forward backtesting, 95% prediction intervals | 2.7% MAPE on synthetic retail data |
| RAG-in-a-Box | Offline retrieval-augmented chatbot with TF-IDF / dense / hybrid retrieval and cited answers | Runs with no API keys |
| Billboard: Top 10 or Not? | Team analysis of 5,342 Hot 100 songs: ten hypotheses and a debut-week hit model | 42.9% precision on unseen songs (random: 5.9%), my extension |
Also: LoRA Lab (LoRA from scratch in NumPy) · Agent Arena (multi-agent debates) · PromptForge (prompt evals) · SplitSmart (AI bill splitting) · LinguistLive (real-time AI voice tutor)
Python · SQL · R · TypeScript · Java — pandas, scikit-learn, statsmodels, PyTorch · React, React Native / Expo, Flask · AWS, Vercel, Docker, Git, GitHub Actions · Tableau, Power BI