Turn existing CCTV into a
multi-vertical analytics product
One edge-first computer-vision pipeline. 19 analytics verticals. Real detection on live public cameras, a model that measurably improves itself, and billing by the camera — all with no raw video ever leaving the site.
Every vertical, one pipeline
The same edge agent runs them all. Live verticals detect real people & vehicles on public cameras today; Simulated ones need a trained model or sensor and show representative data in the demo.
A product that improves itself
Accuracy isn't a marketing claim — it's measured, and it compounds. The accuracy flywheel turns live operation into a continuously better model.
Snapshot
The edge pushes an annotated frame + its predicted count (opt-in, privacy-first — no continuous video).
Label
A human confirms the true count in the dashboard. Every judgement becomes a durable training sample.
Measure
Per-camera, per-vertical accuracy (MAE, % error) is computed from the samples — an honest, auditable number.
Retrain & promote
Once enough labels accrue, a count-gated retrain produces a better model and promotes it per site.
Built like a platform
Not a counting script — an enterprise-ready control plane.
Edge-first privacy
YOLO + tracking run on-site; only counts and events sync out. No facial recognition, anonymous by design.
Live public cameras
Validated end-to-end against real public feeds — same code path as a customer's private camera.
Metered billing
Active camera per month, metered through Stripe. Transparent usage view; no retroactive disputes.
Fleet management
Heartbeats, config push, canary rollouts and device-health alerts — per-camera billing stays operationally sane.
Multi-tenant & RBAC
Row-level tenant isolation, roles (viewer→owner), an append-only audit log, mTLS device identity.
Real-time alerts
PPE, capacity, fire, weapon, fall, intrusion and more — severity-ranked, replayable from the live feed.