1. Business Context
Manual quality inspection created major bottlenecks in circular fashion supply chains, with 30% inconsistency in assessments and excessive labor costs driving up prices by 25%. The industry needed automation to scale sustainable fashion initiatives.
2. Solution Snapshot (24 Months)
| Component | What We Built | Technologies |
|---|---|---|
| Data Infrastructure | Custom dual-camera capture system with specialized annotation platform | Flask APIs, Streamlit UI, RESTful backend |
| Dataset Enhancement | Quality control & balancing for 30,000+ images; metadata enrichment | Custom processing apps, Pytorch |
| CV Model Architecture | State-of-the-art attribute detection with 92% accuracy on primary categories | ConvNeXt, Vision Transformers (ViT) & CLIP variants |
| Synthetic Data Pipeline | Inpainting framework to generate realistic damage patterns for object detection | Prompt engineering, Automated mask generation |
| Deployment System | Web-based interface for real-time attribute detection | Gradio |
3. Impact
- 40% reduction in manual inspection processing time.
- 50%+ decrease in data collection costs through synthetic data generation.
- 92% accuracy on primary damage categories despite challenging long-tailed data.
- Created one of a kind public dataset for fashion attribute detection research.
- Selected as 1 of only 5 projects presented at EU sustainable AI summit (2023).
4. My Role
- Led end-to-end project management for €1M AI initiatives across two major programs (AI for Circular Fashion, CISUTAC).
- Designed technical architecture and coordinated cross-functional implementation.
- Personally implemented key components of synthetic data generation pipeline.
- Delivered presentations at Vinnova Innovation Week (2022) and EU sustainable AI (2023).
5. Next Steps
Read more here.
Need to automate your quality inspection processes using AI? → Book a consultation