Projects
ALPHA-001
Bitcoin Trend Analysis
- Led cross-functional collaboration with S&P Global, translating stakeholder needs into technical user stories to ensure on-time delivery.
- Analyzed large-scale Bitcoin price data with time-series modeling (LSTM/Transformer), identifying the prior 10-hour price as the strongest predictor of short-term trends.
- Applied model interpretability techniques (LIME) to generate actionable insights, enhancing transparency and sponsor confidence in data-driven decisions.
EPSILON-005
Predicting Hypotension with Electrocardiogram
- Built a reinforcement learning model to predict intraoperative hypotension from ECG data, replacing invasive arterial monitoring.
- Developed a custom Gym environment simulating patient variability and stochastic hypotension events.
- Implemented and compared MLP, RNN, and LSTM models; RNN achieved best performance (score ↑ 3.16 → 38.08).
- Designed reward functions and tuned model architecture to improve learning efficiency and prediction accuracy.
BETA-002
UCI Search Engine
- Designed and built a robust search engine in Python with indexing and user interface.
- Developed a scalable, fault-tolerant web crawler for efficient data collection and storage.
- Integrated search engine with OpenAI API for dynamic website summarization.
GAMMA-003
Panorama Mosaics
- Independently completed a Python project for panoramic mosaic generation, manually selecting corresponding points between central and peripheral images.
- Designed and implemented homography transformations for precise image alignment.
- Developed a warping function to align images within the mosaic, interpolating gray values for smooth transitions.
- Employed weighted blending to merge overlapping images seamlessly, ensuring artifact-free results.
DELTA-004
Texture Quilting
- Individually completed a Python project focusing on image texture synthesis and manipulation.
- Implemented the
shortest_pathfunction to find the shortest vertical path through a cost matrix, using dynamic programming and backtracking for accuracy. - Designed the
stitchfunction for blending grayscale images with minimal brightness differences. - Created the
synth_quiltfunction to synthesize textures by stitching texture tiles, resulting in cohesive, visually appealing textures.