LESOTHO · ACTIVE PILOT
LUMA
0 CONVERSATIONS · 30D 0 EMERGENCIES FLAGGED

An AI-native services company replacing the INGO operational layer for community health in sub-Saharan Africa.

Governments contract the service. Pharma F100 buys the data. We do what 1,500-person implementation organizations used to do — at 5–15% of the cost.

luma · MoH Lesotho
● online — replies in Sesotho, English
TRY IT YOURSELF · TALK TO LUMA →
WHY NOW · 2025

The cuts were abrupt. The off-ramp was overdue.

International aid was always supposed to make itself unnecessary. In 2025, US foreign assistance tried to force the issue — leaving ministries with less budget and no implementation partner. Both gaps are real.

50%
RESTORED
ministries spending direct — at reduced budgets
Hover for detail →
20%
DEFUNDED
pure shortfall — services economize or contract
Hover for detail →
30%
INGO LAYER · GONE
the trainers, supervisors, M&E officers — operational capacity gone
Hover for detail →
— Pre-2025 US bilateral health aid in SSA, by what's left of it — 100% baseline · USAID + PEPFAR · 2024
The immediate gap

Ministries are operating. The 50% that's flowing direct gets spent — on the same patients, by the same nurses, in the same districts. They're just stretched in two directions at once: 20% less budget, and no INGO partner left to run the trainings, supervisory visits, M&E systems, or last-mile supply chain that the previous 30% used to deliver.

The bigger argument

International aid has been ducking one question for forty years: the goal of foreign assistance is supposed to be the recipient's eventual independence from it. In practice, off-ramp planning has been politically uncomfortable and structurally absent. The 2025 cuts forced the conversation — abruptly, without warning, with real human cost in the transition. luma sits on the off-ramp side of that argument: software that helps ministries do more with what they have, that scales without a foreign implementing partner, that contributes to the independence aid was always meant to enable.

READ THE FULL THESIS →

WHAT IS LUMA

One platform. Three layers. One destination — ministry independence.

LAYER 01 — THE WORKER

WhatsApp workflow companion in the worker's own language. Surfaces the right protocol, the right contact for escalation, real-time decision support. The thing INGO supervisors used to do, delivered by software that scales.

LAYER 02 — THE GOVERNMENT

Operational dashboard of CHW network activity, severity flags, district coverage gaps, defaulter tracing. Multi-year MoH master service agreement. Sovereign data residency. No INGO middleman, no donor cycle dependency.

LAYER 03 — THE COMMERCIAL WEDGE

The same operational data feeds an API for pharma real-world-evidence buyers. The revenue from that layer subsidises the ministry layer — which means luma is affordable to ministries without donor backing.

The endgame is the same as the long-stated goal of foreign aid itself: a country running its own primary care, on its own schedule, with its own data, without a donor cycle in the middle. luma is the operating system for that destination.

EXPLORE

Every surface is live.

RWE PITCH BETA

Africa is a $250B blind spot. Government-contracted last-mile data is the way in.

10 districts · 3 protocol families · live
Pharma RWE pitch
The commercial layer that funds the independence layer. Trial-site selection, cohort builder, pricing.
Public Health dashboard
Bayesian posteriors against published priors. The surveillance feed that used to come from DHS / PEPFAR partners.
5 indicators · 95% CI
Ministry operational view
CHW network activity, severity flags, district coverage — the dashboard ministries run after the INGO partner is gone.
10 districts · live network
0.4%
HIV PREVALENCE · POSTERIOR
Pharma data dashboard
Treatment cascade, burden posteriors, daily trend. Versioned API.
14:42 Patient missed ART refill — what's the protocol? allowed
14:41 Pregnant woman, severe headache, blurred vision allowed
14:39 What dose of bedaquiline for 45kg patient? refused
14:36 TB contact tracing — exposed family, what next? allowed
Audit log
Every conversation is logged with safety outcome, latency, and retrieved sources.
14,283 entries · 30d
P(θ | data) ∝ P(data | θ) · P(θ)
Projection methodology
How the prior + observed CHW signal collapse into a posterior. Beta-Binomial conjugate update.
1// GET /api/insights/pharma
2{
3  "hiv_prevalence": 0.214,
4  "ci_95": [0.198, 0.231],
5  "sample_n": 14283
6}
API documentation
JSON endpoints for every dashboard. Versioned. Authenticated. Rate-limited.
200 OK
UPTIME · 99.97%
RESPONSE · 0ms
/health
Live status. The endpoint that answers when nothing else will.