Act One

jupyter's development began with a clear architectural decision: edge-first processing. While competitors pursued cloud-based solutions, jupyter recognised three fundamental advantages of local processing: significant reduction in running costs for customers (avoiding expensive cloud API inference costs), sub-100ms response times essential for real-time deterrence, and complete data sovereignty for sensitive security footage.

Our market research and industry insights drive practical solutions that deliver real outcomes.

The core innovation focuses on applying our engine to two areas in the pre-detection and post-detection processing pipelines–– the first applies a set of functions during the face embedding extraction process to simulate the most difficult detection results possible, and the second delivers real-time correction of detection issues caused by camera processing limitations or proximity/skin-tone challenges.

But there's more, the inherent capabilities of this engine determine the skin tone of humans in any footage and dynamically regenerate or resolve tone issues – before sending down the pipeline for analysis and automated security responses.

What started as building a face recognition solution with incredible accuracy results - evolved into something far more powerful. Universal facial recognition accuracy unlocked capabilities previously difficult to attain. For example, rather than creating manual restricted static watch zones for cameras –which can be challenging for mobile security applications––or pan and tilt systems - jupyter's LoiterAI model automatically generates dynamic heat maps for restricted zones using known authorised faces to track and paint restricted zone heatmaps.

Or building real-time Asset Theft Detection that uses both our Face AI system and high-precision posture detection models to establish genuine asset theft intent, before it occurs.

With jupyter's powerful foundation models, existing legacy IP camera networks can be repurposed–– by gaining "intelligent eyesight" through SecureProtect AI. Deployed modern cameras gain even more power. The proposition is compelling and the use cases have real widespread impact, and we are elated to share our offerings to the wider market.

jupyter partnered with Radxa to envision and customise our edge compute hardware––jupyter hub, a small form factor SBC with exceptional AI analysis and GPU processing prowess––we run the SecureProtect workload on a machine that is both right-sized for today, and has the ability to scale and support more powerful use cases in the future.

Constraining the SBC device to a small enclosure meant paying incredible amounts of attention to detail to a device that processes video footage round the clock, generating lots of heat. We always envisaged that customers will mount the hub in server room or data centre locations, so a utilitarian and practical design was crucial.

We also knew that home users may place the products within entertainment unit areas and bookshelves, so it was crucial to achieve a simple and elegant design aesthetics for the market segment too.

We worked with the Design Team at Outerspace Design in Melbourne–– to balance aesthetics, functionality, and practicality for the product, and we are incredibly proud of the design that we've achieved, and we are even more excited to share it with the world.

The most challenging integration aspect proved to be the trifecta of latency, performance, and accuracy under real-world conditions. When your system unlocks garages for registered vehicles, detects package theft attempts in real time, or identifies genuine audio anomalies, there's zero tolerance for errors. False positives create frustration; false negatives create vulnerabilities.

Every component—from facial recognition to audio detection to automated deterrence—required precision engineering for deployment where accuracy isn't just important, but is also critical for user safety and trust.

Solving universal facial recognition didn't just fix one problem—it created a platform where accuracy becomes the foundation for intelligence. When you can reliably identify everyone, you can accurately map behaviour patterns. When face training works in challenging conditions, legacy camera infrastructure becomes intelligent. When detection speed enables real-time response, prevention becomes possible rather than just documentation.

The result: a comprehensive security intelligence platform built on the unshakeable foundation of recognition that works for everyone, everywhere, always.