SafeSpace is shared spatial memory for blind and low-vision people, built on MongoDB Atlas. Wearing Mentra Live glasses, a person asks "Robin, where is the washroom?" or "Robin, what's this?" and hears the answer in about two seconds, in steps and clock positions rather than colours. Staff build the memory just by talking ("extra towels are above the sink") or by walking through a space once: the video is indexed into a searchable map of doors, counters, signs and hazards. We indexed this venue from four phone videos in about two minutes. Every space has its own rules for who sees what. The room guide is shared. A guest's passport, room key or medication stay private to them, expire after the stay, and can be wiped any time with "forget my passport". If someone asks about something they are not allowed to see, the request is denied and logged in an audit trail. A live dashboard shows every memory and every allow or deny as it happens, and changing a rule there changes what each person can retrieve straight away. Safety is built in. Instructions hidden in speech or on signs are refused, card numbers and PINs are removed before anything is stored, and Robin never claims a path is safe or clear. The glasses listen all the time but only answer when addressed as "Robin". Built with a Rust backend, Gemini for extraction, Voyage multimodal embeddings and MongoDB Atlas for memory and search. The same setup works for hotels, offices, homes and venues. About 17 million people in North America live with moderate to severe vision loss or blindness.
Team
