● Designing LLM decoding via multi-agent signaling, improving acc. by 5.2% through self-play.

● Researching on learning natural language algorithms and pseudocode generation.

● Led development of essay grading and staff allocation features in the IELTS system serving 3.5M users.

● Built a BERT model and LangChain pipeline to create essay in-line comments and scores with 83.5% acc.

● Developed an optimized scheduling algorithm and reduced errors by 12%.

● Developed an R&D portfolio optimization model for phase 2-3 drug development.

● Integrated dynamic modeling with optimized penalized regression and decision trees, raising 25% return.

● Data cleaning and testing on pharmaceutical datasets to refine and validate the model.

● Designed and built a tool to automatically code and quantify gaze behaviors in real-world interactions.

● Boosted accuracy by 12% by fine-tuning a Faster R-CNN with an integrated LSTM for gaze detection.

● Developed 3 social interaction tests for assessing ASD and a video data processing pipeline.