NagaTranslate shows the API cost tradeoff in low-resource AI
NagaTranslate is a service for translation and speech work in Nagaland, India, currently covering Nagamese, Ao, and Sema. These languages have mostly been spoken rather than written, so there is little standard side-by-side translation data to train on. Text translation now uses a commercial LLM API with tuned instructions and a few examples.
The project first used a tuned NLLB model, then moved to the LLM API because it gave more natural everyday wording and handled meaning across sentences better. The long-term aim is to move back to self-hosted such as a small Llama or Gemma, so the backend can run independently and avoid ongoing API costs. The main blockers are GPU hosting costs and whether small models can keep enough quality under tight resource limits.
Speech output uses a tuned VITS model trained on custom Nagamese voice data, while uses Whisper tuned on custom Nagamese recordings. The speech pieces run on ZeroGPU behind a secure API layer, and an Android wrapper built with Flutter is in closed testing.
Key points
- NagaTranslate supports Nagamese, Ao, and Sema for translation and speech features.
- Text translation currently depends on a commercial LLM API with tuned prompts and examples.
- The goal is to return to self-hosted to reduce API cost and vendor dependence.
- GPU hosting cost and small-model quality are the main limits.
- Spelling variation creates high token variance, so preprocessing and matter.