Team 17

Steady AI

Steady is an AI shopping agent that helps families make aging parents' homes safer by analyzing room photos to identify evidence-backed fall hazards, researching and evaluating suitable safety products, and preparing purchases for human approval.

Check your own project

Community votes recognize the build and do not affect official judging.

Review the project

Start with the source code, then open the demo or video if available.

View GitHub repository
Project description
Steady is an AI shopping agent that helps families protect aging parents from everyday fall hazards, starting with a photo of a room. 1. Observe and identify: Gemma 4, hosted on Superlinked SIE, describes visible features without making safety claims. GLiNER2.5 Decide classifies those observations against 11 CDC STEADI-inspired fall-hazard categories. Deterministic checks validate the evidence and flag anything the photo cannot establish. 2. Research and evaluate: Confirmed findings become product requirements. Exa searches real product pages; GLiNER evaluates product information, while deterministic rules reject unsuitable options or flag missing specifications for verification. 3. Act with approval: Vercel Workflow SDK coordinates durable steps and pauses for human approval before executing a clearly labeled simulated purchase. 4. Make every decision auditable: SingleStore persists cases, product evidence, decisions, approvals and the audit trail. Hono + Datastar provide a live-updating interface. 5. Test reliability: Firedrill scenarios cover unsafe recommendations, missing evidence, failed searches and attempts to purchase without approval. These scenarios are prepared; live drills remain to be run. Our principle: No unsupported hazard claims. No purchase without human approval.
Tools used