🎓 M.S. Applied Data Analytics — Boston University (GPA 3.75/4.0, graduated May 2026) 🔬 Research Assistant — Quantitative FX & DXY Strategy Research · Prof. Eugene Pinsky, BU 📄 WikiFlow — First-authored paper accepted at EAI CSECS 2026 (Springer LNICST, Scopus-indexed) 📰 Traders Magazine (est. 1982) — Short paper under review, 2026 🏆 MedAI Hackathon 2026 — 3rd of 40+ teams · Amyloid PET Centiloid Prediction 📚 Publications — 2 published (IEEE IATMSI · Springer ICADIE) · 1 accepted (EAI CSECS 2026) · 1 under review 🎓 Teaching Assistant — Advanced Machine Learning & Data Science with Python, BU 🛰️ ML Research Intern — Indian Space Research Organisation (ISRO)
Python · SQL · R · PySpark · PyTorch · TensorFlow/Keras · scikit-learn · XGBoost
GCP (Dataproc · GCS · BigQuery) · AWS (S3 · Glue · Athena · Lambda) · PostgreSQL · Power BI · Tableau
Docker · SHAP · OpenCV · pandas · NumPy
| Project | What it does | Key Result | Stack |
|---|---|---|---|
| WikiFlow | Distributed editor-dropout pipeline · 60M revision events · 1.16M editors · 25 GB on GCP | AUC 0.909 (Elephas/Keras DNN) vs 0.845 baseline · Accepted at EAI CSECS 2026 | PySpark · GCP Dataproc · Elephas/Keras · BigQuery |
| DXY FX Strategies | 24 intraday + momentum strategies · DXY and 6 FX pairs · 15-year backtest | Best: $318.80 from $100 vs $121.72 B&H · Sharpe 1.06 vs 0.22 | Python · pandas · yfinance |
| Hospital Readmission | 30K discharge records · 10 algorithms · threshold tuning for recall | Decision framework + cost-benefit projecting $1.77M annual savings/hospital | scikit-learn · SMOTE · SHAP |
| MedAI Amyloid | Amyloid PET Centiloid prediction · 🏆 3rd of 40+ teams | Test MAE 12.67 CL (r = 0.968) | PyTorch · MedicalNet ResNet-34 · FiLM |
| Drowsiness Detection | 4-stage CV pipeline · RetinaFace + ResNet-18 + Random Forest | ROC-AUC 0.964 · F1 0.923 on NTHU-DDD | PyTorch · OpenCV |
| RetailPulse | 9-table e-commerce star schema · 100K orders · Power BI semantic model | Late deliveries cut review scores 4.29 → 2.57 stars | PostgreSQL · SQL · Power BI · DAX |
| Learning Difficulty | NHIS 2024 public-health survey analysis · 36 models · 9 algorithms | XGBoost best performer | R · caret · XGBoost |
| Jet Engine RUL | NASA C-MAPSS predictive maintenance | AUC 0.945 · R² = 0.50 | R · Python |
| MBTA Community | Boston subway graph analytics | Modularity 0.81 | Python · NetworkX · Folium |
| V2X Network | Wireless V2X performance analysis | R² = 0.869 | R · ggplot2 |
WikiFlow — Distributed ML at Scale First fully cloud-native pipeline for Wikipedia editor-dropout prediction: 60M revision events across 1,158,248 editors (25 GB) on a 1+4 node GCP Dataproc cluster, with 13 leakage-free behavioral features derived from each editor's first ten edits. From-scratch RDD logistic regression (AUC 0.845) and an Elephas-distributed Keras DNN (AUC 0.909). Full run completes in under two hours with no local data movement. Accepted at EAI CSECS 2026 · Springer LNICST.
DXY Intraday Strategy Research Best strategy turned $100 into $318.80 over 15 years (2010–2025) against $121.72 buy-and-hold, at a Sharpe of 1.06 vs 0.22. Tested at 0, 1 and 2 bps transaction costs, with results stress-tested using ANOVA to rule out noise. All figures are from backtesting, not live trading. Traders Magazine — under review.
Hospital Readmission — an honest negative result Ten algorithms benchmarked on 30K discharge records; the AUC came out at 0.5654, barely better than chance. Rather than report the flattering 71% accuracy, I rebuilt the work as a threshold-tunable decision framework with SHAP explanations and cost-benefit modeling at each operating point. The dataset had weak signal; the decision logic was the useful deliverable.
🌐 Portfolio: RyanSingh0.github.io 📧 Email: araj7042@gmail.com 💼 LinkedIn: aryan-meena-32685415a 🔬 ORCID: 0009-0005-6739-8265 📍 Boston, MA — open to any US location + remote 🇺🇸 Authorized to work in the US (F-1 STEM OPT)