I managed to run llama.cpp and NVIDIA Parakeet STT locally on Android for real-time offline translation. Here’s how I built the architecture.

Executive Summary

Hey everyone, I wanted to share a technical breakdown of a side project I’ve been building called Glovira. The goal was to build a voice translator that doesn't rely on cloud APIs so it works entirely offline in dead zones. Getting heavy LLMs to run locally on mobile with acceptable latency was a ni...