2025 07 05 mini dwani v2

Mini - dwani.ai / Cost saving

Run only Gemma3-4b-instruct on llama.cpp with the smallest GPU on GCP

Add ASR + TTS on demand : during events

Conserve the credits- become efficient

Use - free huggingface servers for translation and ASR , run them individually

Skip - tts - find small example for kannada

Run - kokoro / speaxhes.ai and whisper - HF

--

whisper / ast Use indicLid - to identify language and switch between models.

Provide control to user

-

remove - unused subdomain from cloudflare