File size: 2,984 Bytes
e4f7911
879469d
 
e4f7911
 
 
879469d
e4f7911
 
879469d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
58355da
 
879469d
 
 
58355da
 
 
879469d
 
 
58355da
 
 
879469d
 
 
 
58355da
 
 
879469d
 
 
58355da
879469d
58355da
879469d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
58355da
 
 
879469d
 
 
 
 
 
 
 
 
58355da
879469d
 
58355da
879469d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
58355da
 
 
879469d
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
---
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

---