LLM compiler and performance

Developed a mobile application at a startup company that helps users manage personal finance.

Built personalized features based on user data using Go and Flutter with Spanner and gRPC for client-server communications.

Implemented streaming dataflow processor for user transaction data using Apache Beam running on Google Cloud Dataflow.

Analyzed the performances of LSTM and Transformers trained on neural machine translation datasets (WMT19) using BLEU scores.

Researched on properties of model ensemble and knowledge distillation using ResNet on CIFAR datasets with PyTorch.

Ring theory, statistics, and calculus.

Processed genetic sequences with Bash, visualized gene mapping using PCA and dendrogram with Python to prove gene linkage.

Built deep neural networks including LSTM with TensorFlow to transfer knowledge on regulatory elements from human to mouse.

Achieved the best test accuracy of over 80% on mouse data via domain adaptation (adaptive batch normalization, or AdaBN) for models trained on human data (20GB).