ZIVA: Zapdos Industrial Vision Agent

Autonomous AI safety agent monitoring CCTV feeds to detect forklift risks and PPE violations, triggering automatic responses in Slack and SAP.

Comment

Inspi### 💡 Inspiration

Workplace safety in heavy manufacturing and warehousing is still heavily reliant on manual observation or retrospective reviews. Forklift collisions and missing PPE (Personal Equipment) are major causes of workplace injuries and operational downtime. We wanted to build an autonomous vision agent that doesn't just display flashing alerts on a monitor but actively triggers immediate, multi-channel responses (from safety alarms to ERP logs).

🛠️ What it does

ZIVA (Zapdos Industrial Vision Agent) is a continuous perception and execution layer for factory camera feeds.

  • PPE Enforcement: Scans and flags workers missing hard hats/safety vests.
  • Proximity Safety: Dynamically tracks moving forklifts and pedestrians, calculating safety distances in real-time.
  • Autonomous Operations (RPA): When a hazard is detected, the agent immediately executes:
    1. Local Warning Broadcasts (audible siren alarms).
    2. Slack Dispatches (sends immediate incident reports to emergency channels).
    3. SAP S/4HANA Log Synchronization (files compliance records directly into the SAP EH&S database).

⚙️ How we built it

We built ZIVA with a focus on speed, accessibility, and visual excellence:

  • Frontend: HTML5 Canvas rendering engine to simulate high-frequency bounding box coordinates.
  • Styling: Premium glassmorphic interface built using native CSS custom properties, grid layouts, and smooth animations.
  • Logic: Vanilla JavaScript managing the proximity rule engine, warning overrides, and the mock integration layers representing Slack webhooks and SAP API endpoints.

🛑 Challenges we ran into

One key challenge was mapping pixels to real-world spatial distances (feet) accurately on a dynamic canvas. We solved this by implementing a ratio-calibration scale (approx. 30px = 1 foot) which feeds into the agent’s configurable policy engine.

🎉 Accomplishments that we're proud of

  • Creating a stunning, highly responsive dashboard that feels premium, clear, and action-driven.
  • Building a full-loop agent execution path that demonstrates how computer vision can drive instant database updates in S/4HANA.

🧠 What we learned

We learned how important it is to link physical vision metrics to digital workflow systems. A warning flashing on a screen isn't enough; the automation of safety dispatches and compliance logs is what makes an AI agent truly useful in the industrial space.

🚀 What's next for ZIVA

  • Connecting ZIVA to live RTSP streams using a Python backend powered by YOLOv8.
  • Utilizing WebSockets to push live bounding boxes directly to our responsive dashboard in real-time.ration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for ZIVA: Zapdos Industrial Vision Agent

Built With

Share this project:

Updates