Blog Posts

I built an LLM inference server with PyTorch, FastAPI and Hugging Face Transformers, then measured what the KV cache really buys you: 2.2x at 32 tokens, 7.3x at 512.

What I've been working on at Uber Eats, the ML and retrieval ideas I've been learning, and what I want to explore next.

A quick dive into METEOR, a smarter evaluation metric for LLM outputs, and a behind-the-scenes look at how I implemented it for Promptfoo.

How to write QuickSelect & how to avoid worst case time complexity of O(N²)

An introduction to vectors, vector embeddings, and vector databases.

Get a peek into my mind. I'll be writing about personal projects, software engineering experiences, book reviews, and even some controversial history topics like prehistoric India and the Vedic age.