Notes
Writing
Building a 6-Model Ensemble for Smartphone Grading
How we architected a production computer-vision pipeline that combines detection, classification, and segmentation models to grade smartphone condition automatically — and the surprising bottleneck that wasn't the models.
Migrating from Docker Swarm to GKE: 5 Lessons from Production
What I learned moving a multi-model computer vision pipeline from Docker Swarm to Google Kubernetes Engine — including HPA tuning, observability with VictoriaMetrics, and the 'cost-vs-latency' tradeoff that almost broke us.