Back
Context: ProductionRole: Full-Stack Engineer
AutoGrade
Industrial Computer Vision System
Grades palm fruit from three live industrial camera feeds and surfaces results to operators in real time.
Availability
Private production system
- Three concurrent camera feeds
- Three-line operator dashboard
- Offline-tolerant event delivery
About the app
AutoGrade grades palm fruit as it passes three production camera feeds, and puts the result in front of factory operators while the batch is still on the line.
Technology stack
- Python
- FastAPI
- YOLO
- OpenCV
- TensorRT
- Next.js
- PostgreSQL
- Docker
Features
- Live inference on three camera feeds at once, using YOLO, OpenCV, and TensorRT behind a FastAPI service.
- Operator dashboard covering all three production lines, updated live over Server-Sent Events.
- AI capture viewer for reviewing exactly what the model saw on a graded batch.
- Role-based access separating operators from supervisors.
- Offline-tolerant event delivery: local buffering and batch sync with UUID idempotency keys, so a retry never duplicates a reading.
- Weighbridge tickets in four steps with gross, tare, and net weight per truck visit, plus recaps per day and per truck with CSV export.
Responsibilities
- Built the Python inference service around the vision models and the camera pipeline.
- Built the TypeScript API, the event path, and the role-based access rules.
- Built the Next.js operator dashboard and the AI capture viewer.
- Designed the buffering and synchronization strategy for unstable factory connectivity.
- Packaged the stack as versioned containers for deployment inside the plant.