Computer vision · Edge systems
Deploying computer vision systems across physical retail locations
Distributed computer-vision systems connecting cameras, Intel NUC edge devices, AWS pipelines and analytics dashboards for Carrefour, IKEA and Färm.
Full-stack development across web software, local services, deployment automation and production operations.
2019 — 2021
Engineering & field operations
Deployed in production
Production deployments for Carrefour, IKEA and Färm
01 / The problem
Turn in-store video signals into a deployable and operable analytics system across locations with different physical and network constraints.
Who it was for
Retail operators and analytics teams.
Why it was difficult
Diagnosing problems that crossed software, hardware, networking and the physical environment.
02 / My exact role
Full-stack development across web software, local services, deployment automation and production operations.
03 / Constraints
- Heterogeneous cameras and on-site hardware
- Unreliable or restricted network environments
- Repeatable deployments across physical locations
- Software and hardware failure modes
04 / Technical & product decisions
- 01
Treat edge software, device setup, APIs and cloud applications as one operational system.
- 02
Use containerised services and Linux tooling for repeatable deployments.
- 03
Automate new-site installation steps to reduce manual variation.
System view
One product, several connected layers.
Conceptual public diagram — no confidential product detail.
05 / Result & learning
Production systems deployed for Carrefour, IKEA and Färm, later adapted into real-time customer-counting systems during the COVID-19 pandemic.
What I learned
A distributed product is only as useful as its deployment, monitoring and field-recovery workflows.
Technology