Back to Blog
AILLMsPythonDeepfakesContent AuthenticityAPI Deployment

Building an AI Agent That Detects Deepfakes in Kenyan Media: A Python Implementation

9 min read  · 1,762 wordsBy Orandi Felix

The dataset is biased toward Kenyan politicians and urban areas. It doesn’t cover rural communities well. I’m releasing the dataset (with privacy redactions) at [github.com/orandif/kenyan-faces-dataset](https://github.com/orandif/kenyan-faces-dataset) under CC BY-NC-SA.

The false positives were mostly on low-quality phone footage of rural areas. The model tends to flag faces with unclear features as "fake." This is a major limitation for broad deployment.

If you’re in Africa and want to adapt this for another country, I’m happy to help. DM me on Twitter [@orandif](https://twitter.com/orandif) or email orandi@datascience.co.ke.

Share this article: