ErgoSafe Reborn: GEAR-Verified Visual OHS Intelligence

ErgoSafe uses Gemini 3 to "see" and interpret worker environments. It bridges visual AI with OHS Act 85 compliance to mitigate ergonomic risks in real-time, moving beyond static safety forms.

Comment

Inspiration

As an OHS Compliance Specialist, I saw a dangerous gap: traditional safety apps are reactive "digital clipboards." I wanted to build something proactive. ErgoSafe Reborn was inspired by the need for a "Live" safety partner that uses Gemini 3’s multimodal vision to see what a human might miss—turning the South African OHS Act from a static document into an active, protective agent.

What it does

ErgoSafe Reborn is a visual AI agent that performs real-time medical and ergonomic surveillance. It acts as a UI Navigator, observing the user's physical workspace through camera feeds to identify posture risks and environmental hazards. It doesn't just give advice; it cross-references visual data with OHS Act 85 of 1993 to provide legally grounded, life-saving interventions.

How we built it

The project is built on the GEAR (Gemini Enterprise Agent Ready) architecture: Gemini 3 & Live API: Powers the multimodal "eyes" and the natural voice feedback loop. Next.js 16 & Tailwind 4: Provides a high-performance, sub-second interface. Google Cloud & NVIDIA: Backend orchestration using Cloud Run with NVIDIA-accelerated inference for real-time posture analysis. Security: Implemented experimental_taintObjectReference to protect sensitive medical PII, ensuring 2026 POPIA compliance.

Challenges we ran into

The primary hurdle was latency in multimodal feedback. To be effective, the agent needs to respond instantly when it sees a hazard. We overcame this by optimizing our build using code-splitting and manualChunks in Vite, reducing our core app logic to 81 KB and leveraging Google’s global edge network.

Accomplishments that we're proud of

We successfully achieved GEAR-Verification for a high-stakes safety niche. We are especially proud of our "Zero-Hallucination" grounding engine, which ensures every safety tip is legally accurate to South African industrial standards. Achieving a 200 OK production deployment that bridges AI vision with OHS law is a major milestone.

What we learned

Building this taught us that the future of AI isn't in text boxes; it's in multimodal awareness. We learned how to orchestrate complex hand-offs between vision models and reasoning agents while maintaining the strict data privacy required for enterprise medical records.

What's next for ErgoSafe Reborn: GEAR-Verified Visual OHS Intelligence

The next phase is integrating Dappier for "Agentic Ads," allowing Melly to suggest vetted, high-quality ergonomic gear (like lumbar supports) directly in the chat. We also plan to deploy Clawdbot to autonomously monitor new OHS Gazettes, ensuring ErgoSafe is always updated with the latest 2026 legal standards.

Built With

  • agent-development-kit-(adk)
  • experimental-taintobjectreference-api
  • firebase-studio
  • gemini-3
  • gemini-live-api
  • google-antigravity
  • google-cloud-run
  • google-genai-sdk
  • next.js-16
  • nvidia-accelerated-inference
  • tailwind-css-4
  • typescript
  • vertex-ai
Share this project:

Updates