• Built and scaled distributed ML infrastructure to support training, evaluation and batch inference

• Supported distributed execution and scheduling systems to coordinate multi-stage ML workflows

• Collaborated closely with ML researchers and product teams to productionize research workflows

• Built a LLM-based Agentic platform for Frontend UI auto-generation, utilized by 80 teams, handling 17 million executions in 2024.

• Integrated a NLP interpreter to parse Natural & Programming Languages.

• Implemented a CV comparison framework to evaluate and benchmark the generated UI.

• Implemented an embedded warm-up module to preload and orchestrate necessary components, dropping request failures in the first 10 seconds by 74%.

• Refactored hashtag module using gRPC, successfully decreased system latency by 17%.

• Collaborated with multiple teams to develop a multi-recipient log system, distributing messages with Kafka.

• Built a Transformer-based framework for code comment auto-generation, powered by Dual Supervised Learning.

• Developed a type-associated encoder to utilize the pair-wise relationship between Java code tokens.

• Implemented copy mechanism to allow both generating words from vocabulary and source code.

• Introduce a dual supervised learning approach to train two models jointly.