- 🎓 MSc in Data Science @ Nanyang Technological University, Singapore - College of Computing & Data Science (Aug 2026)
- 💼 3 years shipping credit-risk ML end-to-end - Lead ML Engineer @ MyShubhLife, ML & Decision Science @ UGRO Capital
- 🔬 Into GNNs, time-series, model interpretability, efficient ML, and audio - with a soft spot for Kaggle leaderboards 🎹
Languages & Core
ML / Deep Learning
MLOps & Infrastructure
| Competition | What I built |
|---|---|
| 🏈 NFL Big Data Bowl 2026 - Player Trajectory Modeling | Spatio-temporal residual model over a physics baseline (CatBoost + LightGBM + neural head with horizon buckets) on [batch, steps, players, features] tensors - ~0.65 RMSE locally vs. a ~0.70 leaderboard baseline, with a leakage-safe training/eval pipeline. |
| 🧩 NeuroGolf 2026 - Minimal Neural Networks for ARC-AGI | Rule-detection + per-task training pipeline producing the smallest possible ONNX networks that exactly solve abstract-reasoning grid tasks under strict parameter budgets, via automated architecture shrinking and verification. |
| Project | What it does |
|---|---|
| 🎙️ Query by Humming | MFCC + DTW retrieval for humming-to-song search, robust to tempo drift, with alignment visualizations and ranking diagnostics. |
| 🧠 RNN Repair | Influence-style debugging of RNN misclassifications via feature abstraction (PCA + mixture components) and controlled ablations. |
| 🌗 Ray Tracing Engine | From-scratch renderer with a result gallery and benchmarks. |
| 🌀 Multi-Objective Optimization | Python scaffolding for Pareto-front experiments. |
| 🧩 Connectionist Model | Experiments toward global-workspace-style signals. |
