COMPANY
Building the intelligence layer for the physical world.
Organizations have cameras, sensors, and data. What they lack is understanding. SurveiLens exists to close that gap, starting with the environments where the cost of not understanding is highest.
THESIS
We stopped asking the wrong question.
SurveiLens began as a system for detecting threats in video. It worked in the way hackathon projects work, and it failed in the way products fail: too many false alarms, too much latency, too much cost, and an architecture that could never be deployed in the environments that needed it most.
Rebuilding it forced a better question. Not how do we detect threats? but how do we help organizations understand what is happening in their physical environments? Threat detection is one answer to that question. It is not the interesting one.
Atlas is the answer we are building. Security is its first application, not its identity.
HOW WE DEPLOY
We teach Atlas. We do not rebuild it.
We work the way a deployment engineer does: on-site, close to the operation, learning the workflows and the priorities and the things that actually go wrong. What comes out of that is configuration: event definitions, thresholds, escalation paths.
What does not come out of it is a bespoke model for one customer. We hold a hard line here, because it is the line every services-drift company crossed without noticing: if a deployment requires a change to the platform, that change ships to everyone, or it does not ship.
COMMITMENTS
What Atlas will and will not do.
A system that observes physical space owes people a clear account of its limits. These are constraints on the architecture rather than promises about intent.
- 01
Footage stays where it is generated
Perception and reasoning run on hardware inside the building. Atlas does not upload video to us, and it does not send video to a third-party model. This is architectural, not a policy we could quietly change.
- 02
Atlas does not identify people
Tracking assigns temporary identifiers so an event can be described coherently: the same entity, across frames, within a session. Those identifiers are not identities, they do not persist between sessions, and Atlas does not perform facial recognition.
- 03
Customer footage is never pooled
The platform improves through reusable knowledge about environments, event definitions, and deployment patterns. It does not improve by training on one customer's video and selling the result to the next.
TEAM
Who is building this.
A small engineering team. Small enough that the people who designed the perception architecture are the same people who will sit in your emergency department at two in the morning working out what normal looks like there.
That is not a temporary arrangement we are apologising for. For the next several deployments it is the point. The platform only becomes reusable if the people building it have stood in the environments it is built for.