Foundation models and open datasets developed by the AI for Uganda research and engineering team.
Foundation model · JAX and Flax NNX · Apache 2.0
A 500M-parameter instruction-tuned causal language model (ai4ug/pearl-chat-500m-it) engineered in JAX and Flax NNX for conversational Luganda and English dialogue, translation, and reasoning.
Machine translation · PyTorch · 23M parameters
A 23M-parameter Transformer neural machine translation model built from scratch for English-to-Luganda translation with a 6-layer architecture and a 16K-token vocabulary.
Applied machine learning capstones engineered by graduates of our training cohorts.
Computer vision · ResNet-50 · Python
This clinical decision-support application uses a ResNet50-based deep learning model to detect pneumonia in chest X-ray images, applying transfer learning to deliver fast predictions that assist healthcare providers.
Cohort showcase · Open source
Are you a current learner or cohort graduate building an applied AI system for healthcare, agriculture, finance, or Ugandan languages? Submit your repository and live deployment for peer review.
We partner with universities, healthcare institutions, agricultural cooperatives, and technology organizations to build and audit open datasets and localized AI systems.