Mosaic

Mosaic finds problems worth solving, and every way they’ve been approached, including in fields you’d never think to look in. It serves two kinds of people: those who notice a real problem but don’t know how to approach it, and builders who want real problems instead of generic project ideas. Post a problem (or pick one from the feed) and an agent built on Muse Spark 1.3 does the work. It removes personal details and blocks prompt injections, restates the need by its function, then searches three places in parallel: a hand-built casebook, the web for what’s already used in the problem’s own field, and live search in other fields that the model itself chooses. Each candidate is judged for fit, who must act and who it reaches; duplicates are merged; a critic tries to break every cross-field idea; quotes are checked against the actual source page; and a coverage check shows which parts of the need are met and which are still uncovered. The result is one ranked map of approaches, with evidence, limits and honest caveats, not a chatbot answer. Example: for stray dogs hit at night, Mosaic surfaced reflective collars that began as car reflective tape; for an overflowing rooftop water tank, it found the automatic shut-off nozzle from fuel delivery. Answered problems join a shared feed for builders. Every run is traced in an admin view with time, tokens and cost per model call. On 16 test problems with planted look-alike decoys, no decoy reached a result.