Beyond the Bed: Engineering an AI "Steward" for the Remote Work RSI Crisis

Desigan Tharmen — Wed, 04 Mar 2026 00:57:24 +0000

What I Built with Google Gemini

As a safety professional with over 20 years of experience managing compliance for major industrial brands, I recognized a silent, modern crisis: Repetitive Stress Injury (RSI) in remote teams. While office environments are strictly regulated, home offices are often "blind spots" where staff work from beds or sofas, creating massive corporate liability and health risks.

I built Ergo-Safe Reborn, an AI-driven stewardship platform powered by Google Gemini. Unlike a static checklist, Ergo-Safe uses a Multi-Agent Swarm to act as a digital safety officer.

Melly (AI Copilot): Uses Gemini’s reasoning to provide legally grounded advice by ingesting the OHS Act and ISO 45001.

Agentic Stewardship: Using crewAI, I orchestrated a workflow where a "Steward Agent" engages the worker while a "Compliance Agent" verifies the setup against regulatory standards.

Multilingual Escalation: Gemini’s native translation ensures that safety warnings are communicated in the user's native language for zero-error comprehension.

Demo

You can explore the live deployment here: ergo-safe-reborn.vercel.app

What I Learned

LLM Grounding is Non-Negotiable: In safety-critical applications, "close enough" isn't good enough. I learned to use the GEAR Framework to ground Gemini in specific legal texts to prevent hallucinations.

The Shift to AI Orchestration: This project forced me to move from writing line-by-line code to designing "agentic logic." Using Gemini as a pair programmer allowed me to focus on how agents interact rather than syntax.

Soft Skills in Automation: Building a system that monitors human behavior requires empathy. I learned to tune Gemini’s "voice" to be authoritative yet supportive, ensuring staff felt protected rather than policed.

Google Gemini Feedback

The long-context window is Gemini's "superpower." Being able to upload the entire OHS Act as context meant my agents could cite specific regulations with 100% accuracy. The low latency in multilingual processing made real-time safety escalations feel seamless.

The Friction:
I initially struggled with Gemini being too polite. In safety, if someone is at high risk of injury, the tone needs to be firm. I had to implement strict system_instructions to ensure the "Compliance Agent" maintained a high-stakes, professional persona.

I built a Multi-Agent "Stewardship" Swarm to prevent RSI in Remote Teams Remote work is a compliance time-bomb. Employees working from sofas, beds, or non-ergonomic chairs are at high risk for Repetitive Stress Injury (RSI). ergo-safe-reborn.vercel.app