Vault of Time

Lemoha (Sesotho for "to notice"): clinicians' quick voice notes become early signals of illness clusters, so environmental health practitioners know where to look sooner. Notice sooner. When illness has one source, such as a water works, a shop or a school event, each clinic sees only a few cases. The pattern exists days before anyone joins it up. In October 2026, more than 400 people in five neighbouring South African communities fell ill within two weeks, most of them children under four. Officials linked it to the one water treatment works that supplies them. A nurse or doctor records a short voice note the way they already talk, mixed Sesotho and English included. Two Meta models listen at once, and Muse Spark turns what they heard into a case card with no patient names; where it is unsure, the card asks. The clinician checks the card and saves it. The recording is never stored and the transcript is not saved. In the replay, plain statistics compare each area's cards with its usual and raise Watch, Signal or Alert, always with the evidence. The replay's Today screen shows each clinician their own area, and says so when it's quiet. For a signal, Muse Spark ranks possible common sources from what the statistics found. Muse Spark Open, Meta's open-weights model, drafts a notification that a person edits and sends. Lemoha never sends anything itself, gives no medical advice and does not replace notifiable medical condition (NMC) reporting. We tested it on a fictional town where water works A is the planted source, using simulation seeds kept unseen until tuning was frozen. Tested on 10 held-out simulation runs: every run reached Alert (median day 16.5). Muse Spark ranked works A first at the first Alert in 7 of 10, and in all 10 on day 19. No false Signals or Alerts; 7 false Watch signal-days across 20 runs. Built during the hackathon by Nthabiseng Patsa, Vault of Time (Pty) Ltd, South Africa. Made-up data only. Open source under AGPL-3.0. Public replay: lemoha.co.za/demo. Next (not built yet): Join a Network, so Lemoha is useful on an ordinary day, not only when something is wrong. Health professionals join peer networks by role, place or experience, with discussions and curated official guidance. When a member asks "is anyone else seeing this?", the answers can become case cards, and the cards can become signals. No patient details and no AI medical advice. Also not built yet: checking real saved reports for signals, since today the statistics run only on the simulation; risk watch from water notices, sewage spills and heavy rain; Share to WhatsApp; a reminder to notify diagnosed notifiable conditions; closing the loop, so reporters see what their reports led to; and a second scenario for food or poisoning clusters.