Map apps and Street View show you the real place as it is now. Psychologists have studied "mental maps" for decades with pen and paper, but nobody gives you your own. Memory Map compares your drawing of home with the real streets, shows exactly how your memory bent them, and puts your siblings' memories next to yours.
The data it runs on
Your drawing: strokes and labels from the in-app drawing pad, or a photo of a paper sketch.
OpenStreetMap: streets, schools, markets, places of worship and parks around the address you give, via the Overpass API, plus Nominatim to find the neighbourhood.
Matching: your labels are matched to OSM features by name and type (amenity=school, amenity=marketplace, leisure=pitch).
Limits and gaps: small shops, trees and family houses are rarely on the map; streets change over the years; some neighbourhoods are thinly mapped.
Fallback: the user drops a pin for anything unmatched on the real map; unmatched places become "only in your memory" and can be saved to a private memory layer, or added to OSM if they still exist.
1. The mission
Help people see how they carry the places they grew up in, and share those maps with the people who grew up beside them. Every family's neighbourhood, remembered together, before it changes for good.
2. Why people can't resist it
Surprise about yourself. "The school was twice as far as I thought" is a tiny shock everyone wants to feel.
Sibling rivalry, gently. Who remembered it best? Families compare scores in the group chat.
Things only you remember. The mango tree, the shop that closed. Those private landmarks feel precious.
3. How it works for real
Draw: canvas strokes and labels, or a photo read by a vision model (Gemini Flash or GPT-4o mini via OpenRouter) that extracts streets and labelled places.
Find the place: geocode the neighbourhood and pull OSM features within 1.5 km.
Match: a fast cheap model (DeepSeek or Gemini Flash) pairs your labels with OSM features ("Mummy's church" with the nearest place of worship), and a frontier LLM (Claude or GPT) resolves hard cases.
Align: a similarity fit (rotation, scale, shift) using matched places, then a rubber-sheet warp so the drawing can morph onto the real map.
Explain: the LLM turns the leftover errors into plain sentences: distances, directions, bends.
4. Cost per use
Pennies per map: one vision call for photo sketches, one or two cheap LLM calls for matching and explaining, cached OSM queries. Under $0.02 a map.
Free: your own map, accuracy score and distortions. Family ($3 one-off per neighbourhood): unlimited sibling maps, the compare view, a printed poster of everyone's maps layered together.
5. Growth loop
Every map ends with "Ask your siblings to draw it too". Each invite carries the neighbourhood but not your drawing, so they draw blind. Family chats fill with scores and arguments about where the market was. Each Sunday a new Draw a Place prompt arrives for the whole family; every drawing joins your life Atlas, your accuracy trend climbs, and the family leaderboard resets weekly, so people return to keep their streak and their rank.
6. Risks and guardrails
Privacy: the real address is never shown on shared cards; maps are private to the family by default.
Kindness: scores are framed as how memory works, never as failure. Grief and lost homes are handled softly.
Children: no location sharing for current homes of minors; the app is about childhood places, not where people live now.
7. MVP build order
Drawing pad with streets, pins and labels.
Address search plus Overpass fetch of nearby schools, markets, worship and parks.
Label matching and a similarity fit with distortion sentences.
Overlay view with the morph and an accuracy score.