Pascalburume/fuatilia

Fuatilia: a small AI that reads contact tracers' SMS

Fuatilia is a prototype to help a health-zone office follow the contacts of Ebola cases for 21 days. Tracers send one SMS per visit from a basic phone. A small language model on the office computer reads each SMS and quotes its evidence. Fixed rules then route what it reads to a named person. The model proposes, rules route, and people decide.

This is a shared public demo with synthetic data only. Every visitor uses the same copy. It is not a medical device and it never diagnoses. Here the model runs on this server's processor, so reading one SMS takes longer than on the MacBook Pro where it was measured (2–3 s per SMS).

Try it (2 minutes)

1. 1.Open the app in its own tab. Inside this page, some browsers block the cookie that keeps the role you choose.
2. 2.Choose Supervisor (S01), then Open.
3. 3.In the menu, open Simulator. As tracer T02, send C046 vu leo, ana homa, hatapiki (Swahili and French: "C046 seen today, has fever, not vomiting").
4. 4.Watch the office side fill in: stored, read by the local model, checked against the SMS, routed by a rule, owned by a person.
5. 5.Open the new alert: the words the model quoted are highlighted, and a sign the SMS does not mention stays "not mentioned", never "absent".
6. 6.Judge presentation in the menu plays the demo story step by step (its outage step needs a local install). Area map, Evidence and System status show the rest.
7. 7.National view in the menu is the national cloud the office syncs to automatically while online (choose National coordination): 14 zones, this office and 13 synthetic ones, and which zone last synced, and how.

The model file is Pascalburume/fuatilia-medgemma-1.5-4b-gguf, used under the Health AI Developer Foundations terms (see its card). This copy sends no request beyond its own container. The model's file is downloaded once, when the image is built.