Created a chatbot that utilizes state-of-the-art RAG techniques. Included a custom evaluation framework and query routing classification model. Leveraged LLMs for generating synthetic training data and evaluating model outputs.

Created RLHF data across many domains for improving the coding ability of foundation LLMs. Enhanced my understanding of what constitutes effective training data for modern LLMs.

Contributed to vLLM, the state-of-the-art LLM serving library, by designing systems to more efficiently manage GPU resources, writing GPU kernels, and improving the API backend.