State Scale Classification & Triage of Seized and Streamed Media

State Scale Classification & Triage of Seized and Streamed Media

Mission Area

Intelligence - Law Enforcement & National Security


  • There is a mountain of media piling up on investigators
  • Timely insights lead directly to operational impact
  • Failure to deliver on operational priorities has an immediate detrimental effect on national security, prosperity, and the rule of law
  • There are thousands of academic solutions proposed every year in the literature; trialing, evaluating, fine-tuning, integrating, and deploying these is a lengthy technical and commercial process that ensures HMG remains behind the state-of-the-art at all times


Polygeist worked with DASA and MoD partners to build the first version of Ozymandias️®️, a massively scalable multimedia processing system built on high performance computing principles for maximum throughput and minimum latency application of diverse analytics to diverse types of media.

The key principles to make Ozymandias usable in operational contexts were simple:

  1. Ingest any type of media, from any source;
  2. Automatically apply all of the analytics all of the time, to extract all of the signatures;
  3. Cross-match signatures between media, people, devices, and confidential ground truth data;
  4. Provide all of this at national scale, with an intuitive user interface; and
  5. Provide deep observability, robust authentication, fine-grained authorisation, and detailed audit log capabilities.

We achieved this through robust engineering practice, building a foundation on which a variety of use cases could be realised. Analytics can be rapidly developed by Polygeist, select partners, or the customer’s own data science teams, and are deployed on the scalable Ozymandias framework in minutes, not months. The framework ensure that instability or runtime issues in one analytic core are automatically partitioned away and can’t break other cores in the workflow.

A real “can do” attitude; working alongside us as if they were part of our team made it really easy to work with them.

In addition to building the Ozymandias framework, we also developed a method to rapidly and safely pull-through analytics from TRL 3-4 (experimentally demonstrated, usually in a lab) to TRL 7-8 (qualified for live operational use). This method is flexible enough to work with a wide variety of AI and traditional parametric analytics; it allows us to qualify, analyse, fine-tune, and deploy new capabilities through Ozymandias in record time. By integrating these into the Ozymandias framework, all the benefits of the user experience, AAA capabilities, common data model, and observability on the analytic’s runtime performance characteristics are automatically enabled.

Ozymandias scales from a single high-density compute node to a fleet of Kubernetes pods scaled elastically across a ( private or public) cloud estate. It is built for high security environments and the application of highly confidential tools and datasets: it runs as effectively when kept in a locked basement, on-prem, and completely offline, as it does in a virtual private cloud from any major vendor. This flexibility is built-in to maximise the impact on a wide variety of highly confidential missions.

Mission Impact

Through the Ozymandias framework, we have delivered an enormous range of analytics to a number of vital missions, both in the UK and overseas. These confidential tools are used to locate, time-stamp, and attribute images and videos to perpetrators.

When you layer up everything you have there, what you’ve built is digital evidence alchemy

We have uncovered subjects of interest, by linking biometric signatures people in video to illicit media taken in clandestine contexts. We have provided powerful content classification and triage capabilities tailored to meet specific operational objectives.

Analytic Cores

Having all of these analytics available in a single interface has allowed operators to seamlessly pivot from filtering enormous datasets to find, for example, images and videos containing children, to performing biometric matching across their vast data holdings, to a detailed analysis of forensic markers indicating digital deception. This network of intelligence is built up automatically, and is available instantly, providing on-demand self-service insight for investigators and desk officers all accessing a common platform from across the world.