· 1 min read
Zhixuan Tech — AI-Driven Road Paving Quality Monitoring
Built a hybrid edge–cloud workflow for Zhixuan Tech to analyze road-surface imagery in real time. Unified fragmented APIs into a single workflow endpoint, simplified frontend logic by 60%, and delivered sub-2-second operator feedback for AI-assisted quality control.
Context
Zhixuan Tech builds AI-assisted monitoring systems for asphalt paving machines.
Mounted cameras capture surface images in real time; AI models detect defects such as segregation or cracks and guide operators to adjust machine parameters accordingly.
Challenges
- Continuous high-frequency image capture under unstable network conditions.
- Workflow fragmented across multiple APIs — frontend had to coordinate several endpoints per state.
- Need for near-real-time feedback (<2s) with fault tolerance.
- Edge devices required offline operation and deferred cloud synchronization.
Approach
- Designed a hybrid edge–cloud architecture: edge devices queue images locally and synchronize with the cloud for AI inference.
- Refactored backend APIs into a single unified workflow endpoint, reducing frontend complexity.
- Introduced asynchronous job orchestration for image upload, inference, and operator feedback.
- Defined clear workflow states: capturing → analyzing → reviewing → adjusting.
Solution
- Backend built with Go (workflow orchestration) and Python FastAPI (AI inference).
- Implemented message queues and object storage for reliable data flow.
- Simplified frontend logic — one endpoint now retrieves all workflow data per state.
- Designed observability metrics (latency, throughput, failure rate) across edge and cloud layers.
Outcomes
- 🧠 Frontend logic reduced by 60% — fewer endpoints and simpler state transitions.
- ⚙️ Feedback latency <2s, enabling real-time machine adjustments.
- 🛰 Zero data loss under unstable connectivity due to queue-based synchronization.
- 📈 Improved on-site quality control and stronger trust in AI recommendations.
