Hi :>
| Project | Description | Tech |
|---|---|---|
| Efficient GPU NTT | CUDA-based NTT pipeline combining Four-step + Radix-4. Fastest kernel execution among all evaluated implementations (1.53× speedup over baseline) | CUDA, C++ |
| TSP Heuristic | New Kruskal-inspired heuristic for TSP. 88.5% accuracy on 100K-city instance vs 69.1% for MST 2-approximation | C++ |
| Sorting Algorithms | 12 sorting algorithms implemented and benchmarked from scratch across 4 input distributions and sizes up to 10⁶ | C++ |
| Speaker Diarization | Text-based speaker diarization using contrastive learning to learn speaker-discriminative utterance representations | Python, PyTorch |
| CrossFit WOD Analysis | Embedding-based workout clustering pipeline with OpenAI API and structured JSON normalization | Python |
| Contactless Scroller | Hands-free scrolling interface using YOLO-based head-gesture detection and DeepFace expression analysis | Python |
| PintOS Kernel | Implemented scheduling, timer interrupt handling, and system calls in C; debugged synchronization and user-program execution | C |
| Robotics Control Systems | Three control systems in Unity: teleoperation, checkpoint-guided navigation, and LiDAR-based multi-agent exploration | C#, Unity |