How the Platform Is Actually Built

Two sub-platforms, one open-source foundation, and no third-party proprietary code. This page is the technical answer for the people who will have to operate and audit the system.

Book a Demo Platform Documentation

Architecture

Two Layers, on Kubernetes

IREX is a cloud-native aggregation of microservices organized into two sub-platforms. The secure cloud layer runs the data services on 100% open-source software: Cassandra for raw video, Redis for channel status and session cache, Kafka as the message broker between microservices, S3 object storage for snapshots and video chunks, and PostgreSQL for relational data.

The smart-city layer runs the processing on Kubernetes: the per-channel video-analytics modules, the HTTP gateway, the event dispatcher, the face and plate matchers, event search, alarm monitors, authentication, licensing, and PTZ control. Kubernetes provides self-recovery, load balancing, and automatic container migration when a node fails.

An optional edge architecture moves analytics close to the cameras at remote sites, so only events travel to the cloud rather than raw video, with offline recording and automatic event synchronization after the network recovers.

The Pipeline

What Happens to a Camera Stream

  1. 01

    Ingest and Analyze

    A per-camera analytics module ingests the stream, runs the analytics, and generates events with snapshots.

  2. 02

    Dispatch and Persist

    The event interface publishes messages to Kafka and records video chunks to Cassandra; downstream streamers persist full events and event metadata.

  3. 03

    Extract and Match

    Dedicated neural-network services extract face descriptors, detect appearance features, and match detections against the enrolled face and plate databases.

  4. 04

    Index for Search

    Proprietary descriptor clustering and indexing makes billions of face, figure, and vehicle objects searchable in seconds.

  5. 05

    Export and Audit

    Events flow continuously to dashboards, and every high-risk action is written to the append-only audit log against its Case ID.

Availability

The Three Availability Guarantees

  • 100% of system components are hot-swappable, with no single point of failure. Any server, disk, switch, edge node, or analytics worker can fail or be replaced under load without operator-visible downtime.
  • 100% high-load design, with linear scalability across every dimension. Cameras, users, retention period, and active AI modules each scale independently. Doubling one is an operational change, adding nodes, not an architectural one.
  • 100% of platform components monitored 24/7 by the IREX network operations team, on IREX-hosted SaaS and on user-hosted deployments under Premier Support.
  • Storage is triple-replicated with automatic failover, so an individual server or drive failing does not cost you a case.

Facts

Inference
Two interchangeable pipelines: a proprietary CPU framework (Synet) and an NVIDIA CUDA GPU pipeline. Hardware-agnostic; the mix is sized per deployment by IREX engineering.
Model Governance
Every model in the platform is governed by the public AI Model Governance Policy: dataset provenance from four permitted sources, human-reviewed labels, evaluation and bias testing before release, artifact integrity from training run to deployment, and foundation models used as published, pinned by version and checksum and served only inside the instance.
Network
A Leaf-Spine fabric for predictable low-latency communication across the cluster.
Proven Scale
100,000+ cameras on one instance, all running real-time analytics with 30-day instant search.
Architectural Ceiling
1,000,000 cameras, 100,000 users, and a one-year retention period without architectural change.

FAQ

What is the stack, exactly?

Entirely open-source technologies: Kubernetes, Ceph distributed storage, Cassandra, Kafka, PostgreSQL, Elasticsearch, and Apache Spark, on a Leaf-Spine network fabric. No third-party proprietary code is required, which is what makes source review and escrow realistic.

Do we need GPUs?

Not necessarily. Inference runs on two interchangeable pipelines, a proprietary CPU framework and an NVIDIA CUDA GPU pipeline. The platform is hardware-agnostic and IREX engineering sizes the CPU and GPU mix per deployment against camera count, retention, AI workload, and your existing server inventory.

Does multi-agency access work on one deployment?

Yes. The architecture is carrier-grade and multi-tenant with hierarchical role-based access control. Users sit in a hierarchical catalog of groups, each group is granted specific cameras, watchlists, locations, venues, and monitors, and a user can hold a different role in each group. Six roles run from Viewer to System Manager.

How are releases rolled out?

Versioning is MAJOR.MINOR and per-version release notes are published in the platform documentation. Rollout is a managed procedure: IREX DevOps notifies your counterpart of the contents and implications, both sides agree the window, users are warned, and the documented functionality tests are repeated afterwards. Upgrades need no action on the client side.

Send Your Architects

Technical due diligence is welcome, including source review and escrow arrangements for sovereign customers.

Book a Demo Private Cloud