Writing
Notes on image processing and AI systems: experiments, mechanisms, technical decisions and reproducible artifacts.
I distinguish what a source reports, what I interpret and what I have personally reproduced.
Liveness Is Not Model Readiness
What separate liveness, health, and readiness responses mean in a documented vision-model API, and what they do not establish about prediction quality.
5 min read
Silver Labels Are Not Ground Truth: A Four-Head Fashion Classification Project
What do multi-head vision benchmarks measure when annotations come from heuristics and VLMs? This write-up examines the four-head classification setup, reported F1 metrics, and a separate 210-image human label audit.
7 min read
Degenerate Policy in a Highway RL Agent
A PPO agent learned to favor slow driving rather than the intended behavior. This write-up examines the reported training results and reward setup.
6 min read