SafeCo — AI-Powered Community Security
SafeCo turns a neighborhood's distributed evidence into community intelligence. It's not a camera app, a social network or traditional CCTV: neighbors, businesses, cameras, phones, sensors and AI agents collaborate to detect incidents, correlate evidence, raise alerts, investigate events, coordinate the response and build cases — with privacy and human control at the center.
> Value prop: *turn distributed evidence into community intelligence.*
What it is
A multi-tenant SaaS for neighborhood security, Spanish-first, built for Argentina (pilot in Miramar, Buenos Aires) on a Ley 25.326 / AAIP legal baseline. Neighborhoods, HOAs, businesses and municipalities operate on one shared core —with roles, permissions and audit— from a mobile app (Flutter, Android + iOS) and a web admin panel.
Who it's for
Neighborhood associations and gated communities, shops and malls, and municipalities / public-safety teams that today run scattered cameras and WhatsApp groups with no correlation or process.
How it works
Distributed capture (cameras, phones, sensors, wearables) → ingestion → the AI agents filter false alarms, score risk, correlate evidence and build the incident timeline → alerts to the right people → response coordination → a case with traceable evidence. Every sensitive action goes through a human and is audited.
The AI agents
Deterministic compute + natural-language reasoning; they never escalate on their own:
Recognition, False Alarm (cuts the noise), Risk (prioritizes), Alert (who and how), Permission (who sees what), Community, Geo (map and zones), Timeline, Investigation (builds the case), Device (monitors and actuates IoT), Governance (compliance) and Community Health (neighborhood metrics).
Sensitive modules — opt-in, OFF by default
Separable, with consent and legal basis:
- Facial / biometric recognition — optional, OFF by default, opt-in + consent record.
- ANPR (license-plate reading) — in scope and complete, but OFF by default; plate data is personal data (reinforced controls, retention limits, audit). A watchlist match never auto-accuses or auto-escalates — human review.
- IoT / actuation — low-risk actuators first (lights, plugs, sirens); locks/gates out of phase 1. Abstraction layer + adapters (Matter / Home Assistant, local-first). No autonomous AI actuation.
- Emergency Mode — OFF by default; 911 via API or in-app with mandatory human trigger/confirmation; no autonomous escalation.
Evidence
Raw video stored both centralized and edge/on-demand, toggle per source. Chain of custody and audit on every access.
Wearables / Meta Glasses (first slice shipped)
Smart-glasses integration: device registration, consent, commands, capture and notify, with per-device permissions and audit (`wearable-service`, REST API). Mock-first, tested end to end.
Monetization
Multi-payer, mixable in one tenant: neighborhood association, per-household, business and municipal.
Security and privacy
Privacy by default. AI agents never call 911 nor actuate IoT autonomously — always a human trigger/opt-in. Every sensitive action is audited. Multi-tenant with per-organization isolation and RBAC.
Architecture
Modular multi-tenant NestJS backend —ingestion, camera registry, incidents, alerts, community feed, chat, geo-comms hub, identity, Orus agent and wearables— · PostgreSQL · Redis · S3-compatible (MinIO) storage · Flutter app (Android + iOS, ES-first, EN/PT-ready) · React/Next.js admin · LLM routing local Ollama / strong cloud model per task.
Status
The 15 design deliverables are closed; code ships as vertical slices (first slice: Wearables / Meta Glasses, green on tests). Community project, pilot in Miramar (Buenos Aires).