Step Wise

Step Wise helps healthcare professionals review potential fall hazards in an older adult’s home before they leave hospital. Its goal is to turn ordinary phone photos into a practical safety review, connecting what is in the room with the person’s mobility needs. A family member photographs the home. The professional explores a 3D reconstruction, checks potential hazards against the original photos, and decides what needs attention. The intended outcome is a clear, prioritised plan for the family, approved by a professional. Details that photos cannot establish remain flagged for an in-person check. The stack combines Next.js and Three.js for the interface and 3D viewer, with Python and FastAPI for the backend. VGGT estimates camera positions and room geometry, while Gaussian splatting creates the navigable room. The AI pipeline integrates TypeSafe Jev for visual checks and Gemma 4 for explanations and recommendations based on the patient’s needs. We built a real room reconstruction from 39 phone photos and deployed a public demo on Hugging Face. It lets visitors explore the room and inspect photographic evidence for two manually marked hazards. The broader prototype includes professional review and family-report workflows. Our main accomplishment is bringing room reconstruction, visual evidence, and patient context into one application, with the professional retaining the final decision.