Building software for clients since 2010
- 15+
- Years shipping software
- 84%
- Retainer renewal rate
- 4–8
- Weeks to first production ship
AI · Manufacturing
The making of Vision Quality Control
Computer vision defect detection pipeline using YOLO with active learning — achieving 97.3% accuracy on manufacturing line inspection with <200ms inference.
- Client
- Manufacturing Client
- Timeline
- 12 weeks
- Built with
- Python, YOLO, OpenCV, Modal, FastAPI, React
Challenges
- Achieving sub-200ms inference time
- Handling varied lighting conditions
- Building a training pipeline from limited initial data
Solution
We built a computer vision defect detection pipeline using YOLO with active learning for a manufacturing line. The system inspects products in real-time, flags defects, and continuously improves through operator feedback on edge cases.
Services provided
What we built
YOLO-Based Defect Detection
State-of-the-art object detection model trained to identify surface defects, cracks, and anomalies in real time.
Active Learning Feedback Loop
Operators flag edge cases that automatically feed back into training, continuously improving model accuracy.
Real-Time Line-Speed Inspection
Sub-200ms inference per frame keeping pace with production line speeds without creating bottlenecks.
Defect Classification Dashboard
Visual analytics showing defect types, frequencies, trends, and production line performance over time.
Operator Review Interface
Intuitive UI for quality inspectors to review flagged items, confirm or override detections, and add annotations.
MES System Integration
API-based connection with Manufacturing Execution Systems for automated quality reporting and traceability.
Results
97.3%
detection accuracy
Under 200ms inference per frame
50%
reduction in manual inspection costs
