2020–2022
With zero in-house ML expertise, taught myself AI model development and launched a digital-signage ad-effectiveness measurement business from scratch
Background
The company decided to launch a new business measuring digital-signage advertising effectiveness through viewer attributes (age, gender) and viewing rates. No one in-house had machine-learning or IoT experience, so it started from proving technical feasibility.
Constraints
- No in-house ML/IoT knowledge (self-taught technical ramp-up)
- Real-time processing on edge devices
- Privacy by design (attribute estimation, not face identification)
- Diverse installation environments (lighting conditions, camera angles)
Approach
Prove 'can we build it?' as fast as possible. Developed age-estimation and face-attribute models on my own to demonstrate feasibility, then designed everything end to end, from hardware and camera selection to the inference platform. Led ML/IoT validation for a team of seven programmers to drive the business to launch.
Implementation
Built an age-estimation model with PyTorch/OpenVINO, running inference at the edge. Edge computing on AWS IoT Core/Greengrass, with a data collection and analytics pipeline built on Lambda and SageMaker.
Results
- Established the technical foundation of a new business in an organization starting from zero knowledge
- Successful pilots in multiple commercial facilities moved the business toward launch
- Delivered real-time attribute capture and a viewing-rate dashboard as a service
- Led ML/IoT validation for a seven-person team
Learnings
A new venture's early velocity is decided by how fast you can prove technical feasibility. Even without in-house expertise, running the validation loop yourself gets the business off the ground — an experience that now powers my company-wide generative-AI rollouts.