Our RAG pipeline got worse before it got better
Adding a reranker dropped our answer quality. The retriever was not the problem, the evaluation was.
Notes from building intelligent systems.
Practical writing on AI, software engineering, data systems, model architecture, and the measurements that turn an idea into a dependable product.
Linear attention without writing a CUDA kernel
RWKV gets you constant per-token latency and flat memory in pure PyTorch. Here is what it costs you.
- Architecture
- RWKV
- Inference
Two months in Tokyo, on a student visa
An internship at a Japanese bank taught me more about writing code other people can read than any course did.
- Travel
- Work
- Japan