We are sharing something we have been building: SafeCo, an AI-powered community-security platform. Not one more camera, not a social network, not the same old CCTV. The pitch is easy to say and hard to do: turn distributed evidence into community intelligence.
The problem: the evidence exists, but it is scattered
Any neighborhood already has plenty of signal —cameras at homes and shops, phones, sensors, eyes on the street— but it lives in silos, in WhatsApp groups that get lost and in recordings nobody watches until it is too late. SafeCo wires those sources together and adds a layer of AI agents that detect incidents, drop false alarms, correlate what happened and help coordinate the response.
What already works (beta with neighbors)
This is not a mockup. On a multi-tenant Flutter + NestJS stack, we have validated on-device, end to end:
- Panic button and live location —with explicit, time-boxed consent, not permanent tracking.
- "My block": real-time presence of who is nearby and available, without exposing personal data (just a count).
- Live incident map with nearby cameras, where a Geo Agent fuses reports of the same event (same type, within 150 m and 15 min) so one episode does not show up ten times.
- Multi-channel intake: WhatsApp, Telegram and the app feed the same incident pipeline; a model classifies and drafts it.
- Live camera sharing with accept/reject, neighbor chat (group and 1-to-1), and evidence (photo, audio, video) attached to the right incident without duplicating it.
- Block onboarding via invite codes and deep links, and identity with server-verified Google login.
Hands-free: point-of-view evidence with Meta glasses
One of the first modules we built integrates Meta smart glasses: a neighbor can capture hands-free point-of-view (POV) video and photos that flow straight in as incident evidence. The glasses also run on-device hazard detection (optional, opt-in) that suggests —never decides— and always requires human confirmation before an incident is created. With no glasses, the same flow falls back to the phone without losing anything.
Privacy and safety, by design
This is what we care about most, and we locked it as architecture, not as a promise:
- Facial and license-plate recognition (ANPR): optional and off by default, opt-in with a consent record, in separable modules. A match never accuses or escalates on its own: there is always human review.
- The AI never acts alone: it does not call 911 or actuate devices without an explicit human trigger. Every sensitive action is audited.
- Sensitive data stored hash-only by default, on a compliance baseline of Argentina's Ley 25.326 / AAIP (phase 1).
Why we are telling you now
Because we believe in building in the open. SafeCo is in private beta, being tested with real neighbors before we open it up. There is no public download or open pilot yet —when there is, you will read it here first. If you care about the intersection of AI, distributed systems and public safety done with judgment, this is exactly the kind of problem we like to solve.