Real deployment.
Real environments.
Deployed
Field demonstration
Technical delivery
Pilot / integration
Pilot preparation
What has worked, under what conditions, with what limits
EaseHealth
Location
Luweero District, Uganda
System
EaseHealth on Android, adapted MedGemma, offline inference
Scale
Research design
50-CHW cluster-randomised study with IDI Makerere
Approvals
Makerere REC MAKSHSREC-2026-41 · UNCST HS7408ES
Evidence type
Deployment, feasibility, adoption, usability, safety — not clinical effectiveness
The problem
In frontline care environments, clinical teams often need access to reliable decision support while working with limited connectivity and resources.
What the deployment taught us
Offline operation can make advanced model capability available where cloud access is unreliable. Clinical boundaries and refusal behaviour are product requirements. Device capability can become the binding constraint.
A documented moment
During a prolonged delivery, an enrolled midwife used the application to structure the case. The system helps the professional ask the right questions while authority remains with the professional.
Limitations
The deployment does not establish clinical effectiveness or universal accuracy. Evidence is specific to documented sites, participants, device configuration, model version and study design.
Location
Luweero District, Uganda
System
EaseHealth on Android, adapted MedGemma, offline inference
Scale
Research design
50-CHW cluster-randomised study with IDI Makerere
Approvals
Makerere REC MAKSHSREC-2026-41 · UNCST HS7408ES
Evidence type
Deployment, feasibility, adoption, usability, safety — not clinical effectiveness
The problem
In frontline care environments, clinical teams often need access to reliable decision support while working with limited connectivity and resources.
What the deployment taught us
Offline operation can make advanced model capability available where cloud access is unreliable. Clinical boundaries and refusal behaviour are product requirements. Device capability can become the binding constraint.
A documented moment
During a prolonged delivery, an enrolled midwife used the application to structure the case. The system helps the professional ask the right questions while authority remains with the professional.
Limitations
The deployment does not establish clinical effectiveness or universal accuracy. Evidence is specific to documented sites, participants, device configuration, model version and study design.
Location
Luweero District, Uganda
System
EaseHealth on Android, adapted MedGemma, offline inference
Scale
Research design
50-CHW cluster-randomised study with IDI Makerere
Approvals
Makerere REC MAKSHSREC-2026-41 · UNCST HS7408ES
Evidence type
Deployment, feasibility, adoption, usability, safety — not clinical effectiveness
The problem
In frontline care environments, clinical teams often need access to reliable decision support while working with limited connectivity and resources.
What the deployment taught us
Offline operation can make advanced model capability available where cloud access is unreliable. Clinical boundaries and refusal behaviour are product requirements. Device capability can become the binding constraint.
A documented moment
During a prolonged delivery, an enrolled midwife used the application to structure the case. The system helps the professional ask the right questions while authority remains with the professional.
Limitations
The deployment does not establish clinical effectiveness or universal accuracy. Evidence is specific to documented sites, participants, device configuration, model version and study design.
Location
Luweero District, Uganda
System
EaseHealth on Android, adapted MedGemma, offline inference
Scale
Research design
50-CHW cluster-randomised study with IDI Makerere
Approvals
Makerere REC MAKSHSREC-2026-41 · UNCST HS7408ES
Evidence type
Deployment, feasibility, adoption, usability, safety — not clinical effectiveness
The problem
In frontline care environments, clinical teams often need access to reliable decision support while working with limited connectivity and resources.
What the deployment taught us
Offline operation can make advanced model capability available where cloud access is unreliable. Clinical boundaries and refusal behaviour are product requirements. Device capability can become the binding constraint.
A documented moment
During a prolonged delivery, an enrolled midwife used the application to structure the case. The system helps the professional ask the right questions while authority remains with the professional.
Limitations
The deployment does not establish clinical effectiveness or universal accuracy. Evidence is specific to documented sites, participants, device configuration, model version and study design.
Location
Luweero District, Uganda
System
EaseHealth on Android, adapted MedGemma, offline inference
Scale
Research design
50-CHW cluster-randomised study with IDI Makerere
Approvals
Makerere REC MAKSHSREC-2026-41 · UNCST HS7408ES
Evidence type
Deployment, feasibility, adoption, usability, safety — not clinical effectiveness
The problem
In frontline care environments, clinical teams often need access to reliable decision support while working with limited connectivity and resources.
What the deployment taught us
Offline operation can make advanced model capability available where cloud access is unreliable. Clinical boundaries and refusal behaviour are product requirements. Device capability can become the binding constraint.
A documented moment
During a prolonged delivery, an enrolled midwife used the application to structure the case. The system helps the professional ask the right questions while authority remains with the professional.
Limitations
The deployment does not establish clinical effectiveness or universal accuracy. Evidence is specific to documented sites, participants, device configuration, model version and study design.
FIELD EVIDENCE & PIPELINE
Pathways in progress
Phase 1 - Field evidence
01
Offline agricultural AI in Kayebwe
Field demonstration · Phase 1
Lead with Lubaale Lovisa, the phone visibly in airplane mode, and the Luganda interaction. A demonstration, honestly labelled - not a deployment.
02
Foundational literacy and numeracy with Fab Inc
Technical delivery and evaluation · Phase 1
Our educational model work targets Ganda-Gemma and Bridge model variants within a teacher-facing workflow. Rather than assessing hypothetical direct classroom impact, evaluation is focused on PCK (Pedagogical Content Knowledge) and ELL benchmarks.
Phase 2 - Pipeline in progress
03
Agricultural advisory pathways
Pilot / integration work · Phase 2
In partnership with Shamba Records (MOU signed) and Hello Tractor (MOU signed), Crane is investigating integration paths with WeatherNext and Africa's Talking.
04
Voice AI Pathways in Africa
Institutional programme work · Phase 2
Programme
UNDP AI Hub for Sustainable Development - Voice AI Pathways, Italy-India-Kenya trilateral
Crane's role
Lead African technical partner
Announced
GITEX Kenya, May 2026
Partners
UNDP · EkStep/COSS (India) · Msingi AI (Kenya, Sauti ASR/TTS) · Kenyan government channels
Status
Institutional programme work - not a citizen-facing deployment
05
SME work notes and lender pathway
Pilot preparation · Phase 2
Crane is exploring local-language work-note and SME-finance workflows with a UMRA-regulated Ugandan lender. No outcomes claimed.
No outcomes claimed.
06
Device access partnership (Twist)
Pilot preparation · Phase 2
Device capability was identified in our early field trials as the binding limit on scale. Twist has expressed interest in providing refurbished Android devices for offline voice AI deployment. Signed at Bologna, June 2026. No outcomes claimed.
No outcomes claimed.
