Writing
Mostly about evaluation — how you find out whether a system works, and why a single number usually cannot tell you.
When Your Metrics Disagree: Evaluating Saliency Models on ASD and TD Fixation Data
Four standard saliency metrics ranked the same three models differently on ASD and TD fixation maps, and the disagreement turned out to be the finding rather than the noise.
- Research
- Computer Vision
- Evaluation
Shipping RAG That Survives Contact With Real Users
A retrieval-augmented demo and a retrieval-augmented service fail in different ways, and most of the engineering work lives in the gap between them.
- RAG
- Engineering
- Evaluation
What Building sIndeX Taught Me About Handing Over Code
sIndeX started as a script to automate my own repetitive SEO analysis, and the acquisition taught me that undocumented knowledge is a liability with a price attached.
- Building
- Career