C.01 · Concept · Data pipeline
Data pipeline
From a shared post to a pinned place.
Six stages turn a video, Reel or post into place records the atlas can trust: fetch, read, resolve, confirm, pin, plan. Every record keeps the creator, platform and quote it came from.
Flow
One pass, top to bottom · under ten seconds
The user shares a post to trvlr from the source app. The share sheet is the only input surface.
OUT url · media · source app
Resolve the URL, pull caption, creator handle, thumbnails and any location tag. Sample video frames at the shot boundaries.
OUT caption · handle · frames · tags
Text from captions, on-screen text from frames, speech from audio. The output is a list of place mentions with the quote they came from.
OUT mentions[] · quotes[]
Match each mention to a real place: name, category, coordinates, address. Score confidence; anything under the bar waits as unconfirmed.
OUT candidates[] · confidence
The ingest sheet shows what was found. Confirmed candidates are checked; the user saves, unchecks, or dismisses. Nothing is pinned silently.
OUT saved[] · dismissed[]
Saved places join the atlas with attribution attached. Nearby saves chain into a route by proximity and time of day.
OUT place · route
Sources
Five in · one shape out
All sources normalise to the same Post record · platform access and terms to confirm per source
Privacy
Rules the pipeline keeps
No feeds are watched, no accounts are scraped. A post enters the pipeline only through the share sheet.
Creator handle, platform and quote travel with every place and are shown wherever the place is. A feature, not a footnote.
Every place passes the ingest sheet. Dismissals are remembered so the same post is not offered twice.
Removing a place removes its source records. Removing the atlas removes everything, on device and off.