Vijay Daita
I'm a first-year MSCS student at Stanford and an intern at TikTok, where I work on recommendation model serving.
- I recently graduated from UIUC with a BS in CS + Economics.
- Over the summer I interned at Exa AI, improving search quality for recent events and Exa Code.
- As an ECE408 course assistant, I helped teach CUDA concepts, FlashAttention labs, and supported students with debugging and exams.
- At the Gies School of Business, I applied NLP methods to corporate documents and built visualizations and statistical analyses.
- At Mako, I built kernel-generation datasets for PyTorch→Triton and created a synthetic code-data pipeline.
- I optimized speculative decoding for code editing using HuggingFace Transformers and PyTorch, integrating it into a custom VSCode GUI.
- I worked on evaluating LLM long-context code understanding in the RepoQA benchmark.
- I operated LongLakeTech.com, providing frontend development and data science services before selling the website in 2024.
I’m currently interested in making LLM training and inference more efficient for software engineering tasks and ML systems.
Posts
September 12, 2025
Exploring Approximations for BM25
June 22, 2025
Using First Token of a Block to Find Relevant KV Blocks
June 22, 2025
Playing Around With Token Compression
February 15, 2025
Blazing-Fast Code Editing via Multi-Layer Speculation
January 26, 2025
Looking at Linearizing Large Language Models
January 4, 2025
Selecting Blocks for Block-Sparse Attention
June 10, 2024
RepoQA: Evaluating Long-Context Code Understanding