CServe LLM inference serving engine

ESIMD team, part of SYCL compiler for Intel GPU

Did an individual project on mechanistic interpretability of binary and ternary transformer networks.

Supervised by Professor Hongyang Zhang. Worked with PhD Candidate Haochen Sun to develop a GPU-accelerated CUDA zero-knowledge proof framework for deep neural networks and transformers. Published "zkDL: Efficient Zero-Knowledge Proofs of Deep Learning Training" as a preprintPublished "zkLLM: Zero Knowledge Proofs for Large Language Models", accepted at CCS 2024

Co-op with the GPU compiler team

Conducted research to enhance NeRF (Neural Radiance Fields) models for thin objectsSupervised by Amy Tai and Prof. Alex WongPublished "NutritionVerse-Thin: An Optimized Strategy for Enabling Improved Rendering of 3D Thin Food Models" in a workshop at ICCV.

Co-op with the Data, Analytics, Visualization Services branch.