Developed benchmarks for multi-source retrieval-augmented generation (RAG) built via a scalable framework and conducted extensive experiments and ablation studies.

Generated 100k+ synthetic queries for scarce domains to train dense retrievers for the Tip-of-Tongue task and developed new routing and reranking approaches.

Contributed to development and analysis of datasets and ML models used to predict a passerby's intent to interact with a robot and designed human baseline surveys.

Implemented, parallelized, and benchmarked various algorithms (Barnes-Hut, Fast Multipole, etc.) for N-Body simulations.

Developed agents powered by tree search and neural network-learned values to play the board game of Blokus Duo.

Finetuned Mistral 7B-Instruct on an augmented ECTSum dataset created with Gemini Pro to improve financial document summarization peformance.