Koshee CEO Jonathan Edmondson sat down with The Uncaged Show to discuss why most AI fails in the field — and what it takes to build systems that actually work.
Our CEO Jonathan Edmondson recently joined The Uncaged Show to talk about what Koshee has learned from four years of research into real-world computer vision, why specialized models consistently outperform generic ones, and the problem that originally motivated the company. It’s worth a listen.
Highlights from the conversation
On why generic AI models fall short in security environments
Jonathan explains why our research led us away from large generic classifiers and toward specialized detection models — and why that decision matters for accuracy, speed, and real-world adoption. When you’re processing 30 frames per second across 1,000 cameras, model size isn’t a virtue.
On processing 10 days of footage in 50 minutes
Law enforcement can’t manually review weeks of CCTV footage after an incident. Jonathan walks through how our Tracker technology processed 10 days of footage in under an hour — generating over 2 million accurate object detections — and how investigators can search that database from a single reference photo, even when the subject is masked.
On the human side of AI adoption
Jonathan pushes back on the narrative that AI is about replacing people. His view is technology succeeds when it makes teams more effective, not when it removes them from the equation.
“You can’t automate accountability.”
— Jonathan Edmondson, CEO, Koshee Protect
On why they built Koshee in the first place
Jonathan and James are from Nashville, Tennessee. The school shootings that have hit their community are a direct part of why we built Koshee. Jonathan talks about that origin honestly — and about what it means to build technology that addresses a problem that shouldn’t still be unsolved.
Listen to the full episode
The full conversation covers our approach to systematic machine learning, the technical decisions behind building on-prem AI, and what real enterprise adoption of AI actually requires. Find it wherever you listen to podcasts, or watch below.