Data Science & Business Analytics student at the University of London (academic direction of LSE, delivered via BIBF), on a full academic scholarship, alongside a Full-Stack Developer Diploma at Reboot Coding Institute.
My focus is machine learning and data analytics, with an emphasis on rigor and reproducibility: leakage-free pipelines, verified explanations, and documented evaluation. Based in Bahrain.
- Interested in data science, machine learning, and financial services & analytics
- Projects spanning credit risk, quantitative finance, NLP, computer vision, and recommender systems
- Reach me on LinkedIn
| Project | What it does | Result | Stack |
|---|---|---|---|
| 13f-portfolio-analysis | Reconstructs institutional portfolios from raw SEC EDGAR 13F filings, with per-filing value-unit verification and amendment-aware diffs | 99.68% of value resolved to tickers | Python, pandas |
| bellwether-agent | Autonomous agent over the 13F database that ranks portfolio changes, investigates them with tools, and writes sourced memos where every figure traces to a tool result | 60% task completion vs 30% baseline (30 trials) | Python, Groq, SQLite |
| credit-scoring | Explainable probability-of-default model on the Home Credit dataset, with verified per-client SHAP explanations | Test AUC 0.77 | Python, LightGBM, SHAP |
| guidely | Internal knowledge assistant (RAG) that answers questions on company documents and cites its sources | 100% retrieval@3 | FastAPI, FAISS, React, Llama 3.1 |
| vision-track | Real-time multi-stream person detection, tracking, and ROI counting with a live dashboard | Precision 0.991, 1.55x CPU speedup | PyTorch, YOLOv8, ByteTrack, ONNX |
| document-categorization | Multilingual NLP system classifying documents across four languages into 7 topics, with context-aware tag extraction | 83.95% test accuracy | TensorFlow, DistilBERT, spaCy |
| sp500-strategies | Leakage-free ML trading signal on S&P 500 constituents, backtested as a long/short strategy | 2-3x lower volatility than the index | pandas, scikit-learn, ta |
| matrix-factorization | Movie recommender on MovieLens 1M comparing SVD and from-scratch PMF, with a Streamlit dashboard | PMF test RMSE 0.8499 | NumPy, SciPy, Streamlit |
Python (pandas, scikit-learn, LightGBM, PyTorch, TensorFlow/Keras, OpenCV, SHAP) - NLP & RAG (DistilBERT, spaCy, FAISS, Llama 3.1) - Computer Vision (YOLOv8, ByteTrack, ONNX) - JavaScript (React) - SQL - FastAPI - Streamlit - Go