Spaces:
No application file
title: Swahili Developer AI
emoji: 🌍
colorFrom: red
colorTo: blue
sdk: gradio
pinned: true
🌍 Swahili Developer AI
We’re a small but growing effort focused on building AI that actually makes sense in African contexts.
It started with Swahili, but it’s not just about one language. We care about how AI works across African languages, systems, and real-world environments — especially where data is limited and infrastructure isn’t perfect.
What we’re trying to do
Most AI today is built somewhere else, for somewhere else.
We’re interested in flipping that — building models and tools that:
- understand local languages (not just translate them)
- work in low-resource settings
- can actually be used in sectors like healthcare and education
What we work on
Language models
We experiment with fine-tuning and adapting LLMs for:
- Swahili and other African languages
- Instruction-following and reasoning
- Domain-specific use (like healthcare)
Multimodal AI
Not everything is text. We’re also exploring:
- Medical images (like X-rays)
- Document understanding
- Combining text + vision in practical ways
Healthcare AI
A big area of interest for us:
- Clinical decision support
- Simple, reliable tools for frontline health workers
- Models that can run even with limited connectivity
Data
A lot of the challenge is data. We spend time on:
- Building and cleaning datasets
- Working with low-resource and noisy data
- Figuring out what “good evaluation” looks like outside English benchmarks
Running models locally
We care about making things usable:
- Smaller, efficient models
- LoRA / quantization
- Running models on limited hardware
Projects (ongoing)
- MedAI Africa – exploring multimodal AI for clinical support
- African Instruction Data – building datasets for training and evaluation
- Local LLM workflows – simple pipelines for fine-tuning and deployment
How we think about this
We’re not trying to chase hype.
We’re more interested in:
- What works in practice
- What can be deployed
- What actually helps someone on the ground
AI doesn’t need to be bigger — it needs to be more relevant.
Collaboration
If you’re working on similar problems — especially around:
- African languages
- Healthcare AI
- low-resource ML
We're open to collaborating.
Tech we use
Mostly practical stuff:
- Transformers, PEFT, LoRA
- PyTorch / PyTorch Lightning
- FAISS for retrieval
- FastAPI, Gradio
- some ONNX / TensorRT when needed
Long-term
We’d like to see:
- Stronger datasets from Africa
- Models that understand context, not just language
- More local ownership of AI systems
Follow / Contribute
If this resonates, feel free to follow or contribute.
Let’s build things that actually work.
👉 https://huggingface.co/swahilidevelopers