Building African-led AI for better healthcare.
CHAIRII is a health artificial intelligence research, innovation and implementation centre based in Nairobi, Kenya. We develop, evaluate and deploy clinically grounded, locally relevant AI systems for African health systems and other resource-constrained settings.
Our work spans clinical decision support, speech and language AI, medical computer vision, health datasets, AI evaluation, responsible AI and implementation science.
We believe the next generation of Health AI for Africa should not simply be imported. It should be built with African data, African languages, African clinicians and African health systems in mind.
| Area | What we are building |
|---|---|
| π§ Clinical Decision Support | Bespoke clinical AI models combining machine learning, compact language models, guideline logic and structured clinical knowledge. |
| ποΈ Speech & Language AI | African-language ASR, clinical transcription, ambient documentation and voice-enabled clinical AI. |
| ποΈ Medical Computer Vision | Vision models for diagnostic and screening applications including tuberculosis microscopy and retinal imaging. |
| π Health AI Datasets | Clinically validated datasets, benchmarks and evaluation resources representing African populations, languages and healthcare workflows. |
| βοΈ Responsible Health AI | Safety, ethics, governance, bias evaluation, clinical validation and regulatory readiness for AI used in healthcare. |
| π₯ Implementation Science | Translating AI prototypes into tools that can function in real clinical workflows, including low-connectivity and resource-constrained environments. |
SikizaCare β from sikiza, βto listenβ β is CHAIRII's programme for developing locally adapted automatic speech recognition and voice AI for healthcare.
The programme is initially focused on Swahili and Kikuyu, with expansion into other Kenyan and African languages.
Our work includes fine-tuning and evaluating open speech architectures including Whisper and Meta MMS, developing medical vocabulary and contextual correction pipelines, and creating clinician-validated speech datasets.
SikizaCare is designed to support:
Clinical transcription Β· Ambient documentation Β· Telemedicine Β· Medical dictation Β· Voice-enabled clinical decision support
Our longer-term objective is to combine speech recognition, clinical reasoning and decision support into a unified African clinical voice AI stack.
TAMBUA-TB explores how computer vision can improve tuberculosis diagnosis using AI-assisted microscopy.
The project focuses on building vision models capable of identifying acid-fast bacilli from Ziehl-Neelsen microscopy images and supporting outputs aligned with routine TB microscopy workflows.
The broader objective is to make AI-assisted TB screening accessible in resource-constrained laboratories using low-cost imaging and smartphone-assisted microscopy.
The work combines:
Computer vision Β· Microscopy Β· Diagnostic AI Β· Smartphone imaging Β· Clinical validation
CHAIRII is developing computer vision approaches for Retinopathy of Prematurity (ROP) screening, an important cause of preventable childhood blindness.
Our research includes retinal vessel segmentation and methods for reducing the dependency on expensive expert-labelled datasets.
One strand of this work has evaluated Frangi-Hessian pseudo-labelling for retinal vessel segmentation, combining classical image-processing approaches with deep learning to support scalable development of ROP screening models.
The goal is to develop clinically useful AI approaches that can expand access to ROP screening in settings where specialist ophthalmology capacity is limited.
Chunguza is CHAIRII's work on AI-supported cervical cancer screening and clinical decision support.
The programme explores how structured clinical guidelines, machine learning and AI can support healthcare workers through screening, triage, referral and follow-up pathways.
Our approach emphasizes guideline fidelity, explainability, safety and integration into routine clinical workflows, rather than treating the language model itself as the clinical guideline.
CHAIRII is developing locally adapted ambient-scribe and clinical-language technologies designed for African healthcare environments.
Through clinical and telemedicine programmes, we are exploring the creation of multi-turn clinicianβpatient datasets representing real healthcare workflows and local language use.
Priority clinical domains include:
Maternal & child health Β· Vaccination Β· Outpatient care Β· NCDs Β· Diabetes Β· Primary healthcare
The programme connects three layers:
Speech β Clinical documentation β Clinical decision support
Our objective is to reduce documentation burden while creating the infrastructure needed for locally relevant clinical AI.
High-performing Health AI requires representative data.
CHAIRII is developing approaches for responsibly creating and validating African health datasets, particularly in areas where global AI datasets remain limited.
This includes:
African-language clinical speech datasets
Multi-turn healthcare conversations
Medical terminology and contextual vocabulary
Clinical decision-support benchmarks
Medical image datasets
Model safety and performance evaluation datasets
Where appropriate, datasets are developed together with clinicians, health institutions, professional associations and research partners.
We believe open science can accelerate Health AI innovation across Africa.
CHAIRII is building an open-to-deployment pathway in which selected research models, benchmarks, datasets, evaluation tools and early-generation technologies can be released under appropriate open licences.
This allows researchers and developers to:
Reproduce our research β benchmark new models β fine-tune locally β identify limitations β build better African Health AI systems
Where technologies progress toward regulated clinical deployment or contain protected clinical data or intellectual property, separate controlled or commercial versions may be developed.
Models must solve genuine clinical and health-system problems.
African languages, workflows, infrastructure and patient populations should be represented from the beginning.
Not every healthcare problem requires a frontier-scale generative model. We explore classifiers, specialised models, compact language models and deterministic clinical logic where these approaches are safer and more efficient.
Clinical AI requires benchmarking, safety testing, external validation and continuous monitoring.
We support interoperable approaches including FHIR, SMART Guidelines, structured clinical knowledge and standards-based clinical decision support.
AI should strengthen clinical decision-making rather than silently replace healthcare professionals.
This organisation will host CHAIRII's growing portfolio of:
| Repository type | Examples |
|---|---|
| π€ Models | Clinical AI, ASR, vision and specialised healthcare models |
| π Datasets | African clinical language, imaging and benchmark datasets |
| π§ͺ Benchmarks | Clinical safety, accuracy and localisation evaluations |
| π€ Spaces | Interactive demonstrations of CHAIRII technologies |
| π οΈ Tools | Training, evaluation, clinical knowledge and data-curation utilities |
Repositories will be accompanied by documentation describing intended use, limitations, evaluation methodology, licensing and responsible-use considerations.
We collaborate with health ministries, hospitals, universities, researchers, professional associations, technology organisations and global health partners working to advance responsible AI for health.
We are particularly interested in collaborations involving:
African clinical datasets Β· African-language AI Β· Clinical decision support Β· Medical computer vision Β· AI evaluation Β· Responsible AI Β· Digital health interoperability Β· Health AI implementation
Centre for Health AI Research, Innovation & Implementation
π Nairobi, Kenya
π African-led. Clinically grounded. Open where possible. Built for real-world health systems.
Research Β· Innovation Β· Implementation