~/ now

Now

Updated October 2026

• Building a ~125M parameter code LLM from scratch in plain PyTorch, one concept per stage, from tokenizer to serving.
• Contributing to vLLM and digging into serving internals: scheduling, KV cache, and batching.
• Shipped RootGraph at the Graph Hacks hackathon (FalkorDB × WeMakeDevs): graph context for agents on a small offline Qwen3.
• Building KJ-Mini, a 3D-printed wheel-legged penguin robot.

This is a now page.