Blog

Here I share my thoughts, learnings, and experiences.

The Smallest Model I've Ever Shipped

July 16, 2026

Embedding a 1.2B regional-language model next to a frontier API: why the UI is the model card, what a 1,024-token context forces on your architecture, and serving a model with two caches.

Understanding KV Cache Bugs in Transformers

January 27, 2026

A deep dive into subtle bugs in KV cache implementations that cause gibberish outputs during text generation, and how to fix them.