Jan 2024 – May 2025 · 1 yr 4 mos · San Francisco Bay Area · Hybrid

• Worked on deep learning research team, focussed on LLM training and inference.
• Built production LLM inference stack based on TensorRT-LLM + Triton Inference Server + gRPC client.
• Trained SOTA Medusa-1 adapter for Llama-3.1 8B, using Liger kernels for training and SGLang Runtime for self-distillation.
• Trained and deployed Eagle adapters for speculative decoding for Llama-3.1 models. Wrote fused linear soft-target cross entropy Triton kernel for training models on A100s.

May 2023 – Aug 2023 · 3 mos · San Francisco Bay Area · Hybrid

• Worked on Snowpark client library, which provides pythonic PySpark-like dataframe APIs by translating dataframe manipulations into SnowSQL.
• Received full time offer.

Jan 2023 – Dec 2023 · 11 mos · New York City Metropolitan Area

• Worked on sequence-parallel QLoRA framework to efficiently fine-tune LLMs for long context tasks.

Jul 2020 – Jun 2022 · 1 yr 11 mos · Bengaluru, Karnataka, India

• Worked on Data Lake, an internal Big Data product which hosts >20 PBs of firm data.