Project title (Company / Domain)
Context: constraints, team size, timeline.
Outcome
- Latency: 120ms → 38ms
- Cost: −42% cloud spend
- Accuracy: +6.2 mAP
What I did
- Data: curation, augmentation for rare events
- Training: loss tweaks, mixed precision
- Optimization: QAT, TensorRT, ONNX
- Deployment: containers, monitoring, rollbacks
Stack
PyTorch, TensorRT, ONNX, Jetson Orin NX, Triton, Grafana
“Short testimonial or anonymized Slack note.”