Soma Intelligence
Computer vision that turns existing security cameras into operational metrics
Role: Founder & Developer
Rol: Fundador y Desarrollador
Overview
A platform that plugs into a venue's existing IP cameras and turns the video into live occupancy metrics and security alerts. A mini-PC on site runs the vision models and emits only JSON events — the raw video never leaves the venue, and there is no facial recognition, just anonymous silhouettes. Privacy is the feature, not a disclaimer.
The Challenge
Venues already have cameras, but the footage is only ever watched after something goes wrong. Turning it into live operational data means running vision models cheaply on site, and doing it without shipping customers' raw video to a cloud they don't control.
The Approach
Four components: a Python edge service doing detection and tracking, a Spring Boot backend that aggregates events into metrics and evaluates alert rules, a React dashboard, and an MQTT broker as the edge-to-cloud contract. The whole stack — Timescale, EMQX, MinIO, API and web — runs as containers with a fail-fast production profile.
The Outcome
The cloud side is deployed end to end behind its own domain, with signup, venue setup, the event pipeline and the dashboard working. The initial vertical is hospitality; the edge service is the piece still being hardened.