DeepStream Audience Detection Pilot

The problem I addressed

I worked on a pilot that needed a practical way to process live video and help an advertising team evaluate billboard audience activity for out-of-home media reporting.

What I built

• A real-time video-processing pipeline using NVIDIA DeepStream
• TensorFlow-based computer-vision processing
• LLM-assisted object classification within the analysis workflow
• A pilot output designed for an advertising agency and OOH media buyers

My role

I built the technical pipeline and support its live commercial pilot.

Why this matters to me

I use this project as evidence of current computer-vision and systems-integration work. I keep it below my core web and product applications because it supports, rather than defines, my current portfolio direction.

Decisions I made

• Used a streaming pipeline suited to real-time video workloads
• Kept machine-vision processing separate from the classification layer
• Kept client footage, pilot data and performance claims private until they can be shared responsibly

Challenges I handled

• Coordinating real-time video and model-processing stages
• Producing useful output without overstating pilot results
• Protecting commercial footage and audience data

System signals

• Live commercial pilot
• Real-time pipeline
• Computer vision
• Dubai agency