Why I Built Dawida
Kidaw'ida (Taita) has minimal formal documentation. Most of it lives in the memory of older speakers rather than anywhere written down and if nobody writes it down, a lot of it disappears with them.
Dawida is a mobile app I started building to change that: a place for the Taita speaking community to document their own language, together.
1. How It Works
Contributors add words with meanings in English or Kiswahili, example sentences and pronunciation audio. They can also share culturally significant photos with bilingual captions and an audio caption. Other members verify entries through voting, so the lexicon that builds up is checked by the community rather than just self reported.
Beyond straightforward word entry, Dawida includes:
- Discover, a daily pick that surfaces a word or entry to the whole community.
- A language health dashboard showing word, example and recording counts, how many entries are verified and how many contributors are active.
- A personal contribution history and streak, so people can see their own impact over time.
- Content flagging, so bad audio or incorrect entries can be reported and reviewed.
2. What's Actually Running Today
The app is built in Flutter and is live, with sign in through email and password including email verification and password reset.
The backend it runs on today is the original serverless one. In parallel, I'm rewriting it into a proper layered API in Go and PostgreSQL, using Chi, sqlc, MinIO for audio and photo storage and JWT and Google sign in for auth. That rewrite is tested end to end locally through Docker Compose, but it isn't yet what powers the live app. I want to be upfront about that distinction rather than implying the new backend is already in production.
Backend repository: github.com/jomboi8/davida-app-backend App: Play Store
3. What Dawida Is Not, Yet
Dawida is an app for curating digital language resources. It doesn't contain a trained translation model or speech recognition. What it does produce, a growing, structured and community verified lexicon of words, translations, audio and examples, is exactly the kind of resource that low resource NLP work like machine translation or speech recognition actually needs before any of that becomes possible.
