# Evarest Technologies

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Stella Oiro](https://cerebralvalley.ai/u/SStellaOiro)
- **GitHub:** https://github.com/EvarestTechnologies/lipwad
- **Demo video:** https://www.youtube.com/watch?v=nn-8eTlnbxk
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2886

Lipwad gets Kenyan hospitals paid.

In Kenya a patient pays a hospital in one of three ways: cash, private insurance, or the Social Health Authority (SHA), the national health insurance. With SHA the hospital treats first and claims later, and the money is lost over paperwork: an unsigned form, an invoice that isn't itemised, a practitioner missing from the registry, a date that doesn't match. SHA's own figures show about 1 in 5 county hospital claims go unpaid, 91,935 claims were rejected for missing or incomplete documents, and 23,000 more because they were not resubmitted in time (SHA figures reported by The Star, Feb and Jul 2026). SHA screens claims with automation; hospitals have nothing on their side.

Lipwad is a multi-step agent for the hospital claims clerk. It reads photographed claim forms and invoices with Muse Spark, and listens to staff voice notes with Muse Voice Transcribe. Every extracted value carries a citation: the region of the photo, or the exact seconds of audio it came from. The agent then checks the practitioner and facility against the registry, runs SHA rule checks, cross-checks the paper against the voice note, and decides green (ready), amber (one named fix) or red (blocked), with reasons. Muse Spark Open explains the outcome in English and Kiswahili, but the rules set the status, so the model can never turn a blocked claim green. When SHA returns a claim, Lipwad reads the corrected page, shows exactly which field changed and where it came from, and resubmits only that correction, with the deadline countdown in view.

Lipwad is not medical AI. It never gives clinical advice, never adds or upgrades services or amounts, and nothing reaches SHA without a named person's confirmation. A blurred photo or an unclear voice note produces a retake request, never a guess.

Live demo (synthetic data only): https://stellaacharoiro--lipwad-web.us-east.modal.run

Built solo by a clinical officer and founder. On 12 synthetic cases covering clean claims, eight common failure modes and a rejected claim, the live agent reaches the expected status and reason in every case, in 22 to 49 seconds per claim. 253 automated tests run in CI against recorded model responses, so CI never calls a paid model.

## More from Meta Global AI Developer Hackathon

- [TraceGaze + Muse](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2883)
- [Excalibur](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2884)
- [khaled ehealth](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2885)
- [Koblip](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2887)
- [precious](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2888)
- [Evangelina Hoepfer Tinkers Space](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2889)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2886. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt