EcoFlow AI

View on GitHub

Digital twin platform for venues that tracks crowd movement and adjusts HVAC and lighting in real time — balancing energy use with attendee comfort.

EcoFlow AI live monitor flagging overcrowding at 120% occupancy in a busy station concourse
EcoFlow AI live updates feed showing people counts and zone occupancy
EcoFlow AI monitoring a live camera feed of a crowded concourse

Built with my team for the Google Intelligent Planet Hackathon at KFUPM, where we were global finalists: top 25 of 2,000+ teams, top 10 internationally, and the only African team in the finals. As the team's backend and cloud engineer, I designed the Google Cloud architecture around a split-stream Pub/Sub pipeline that ingests live environmental and occupancy data. A hot path feeds Firestore for the live dashboard, and a cold path lands in BigQuery for historical analytics and model training. On top of that pipeline, YOLOv8 handles low-latency people detection and Gemini reasons over occupancy patterns to decide HVAC and lighting adjustments. I built the backend services and deployed them as containerized, auto-scaling microservices on Cloud Run, sized to absorb traffic spikes during large events.

Technologies

  • Pub/Sub
  • Firestore
  • BigQuery
  • Cloud Run
  • Docker
  • YOLOv8
  • Gemini
  • Python
  • Django
  • PostgreSQL
Next