Real-time AI inspection for medical device cleanliness
Inspectors reviewed borescope video by eye — slow, subjective and almost impossible to audit. We trained and optimized vision models for in-browser inference and wired them into the operator workflow.
95% detection accuracy in production
30% faster processing
Real-time inference
What was actually broken
Verifying that surgical instruments are free of debris and defects depended on human judgement over long video feeds. Throughput suffered, results varied between inspectors, and there was no structured record to show regulators.
- Client
- A US medical device manufacturer
- Industry
- Medical devices
- Duration
- Ongoing engagement
- Team
- Senior engineer-led
- Stack
- TensorFlowOpenCVAngularDjangoPostgreSQLAWS
The approach
- Trained and tuned computer-vision models to detect defects and debris in real time
- Optimized model inference for the web so operators see results instantly
- Built an Angular interface for live image display and inspection decisions
- Backed it with Django REST services and PostgreSQL for scalable data handling
- Deployed on AWS (EC2, S3, Lambda) for high-throughput image processing
The results
Want references from this sector?
We can introduce you to the client team behind this engagement, subject to their availability. Ask us on the first call.
Other engagements
Let's talk about your version of this
Bring the constraint, the deadline and the system you inherited. We will tell you what is realistic before you commit to a budget.