Every Tongue: AI Scripture Study in Any Language

Every Tongue turns a Scripture passage into grounded study material, English back-translation, review notes, and a reviewer handout. It does not translate Scripture. Instead, the app pulls both English and target-language Scripture from ScriptureFlow, then prompts Google Gemma with expert human-translated parallel text. This grounding keeps devotionals, summaries, questions, and teaching notes tied to real Scripture rather than relying on the model’s memory. Over 7,000 languages are spoken worldwide, but most AI tools serve only a small number well. Pastors, missionaries, translators, and ministry workers in underserved communities often lack safe, reviewable Bible study materials in the languages their people actually use. Every Tongue closes that gap. AMD compute is central. We ran Gemma 3 12B on AMD GPU hardware using ROCm and vLLM, with benchmark results, VRAM usage, and multilingual sample outputs documented in the repository. The app also labels backend mode so judges can distinguish real AMD/Gemma inference, Fireworks fallback, and mock preview behavior. Our testing revealed the key insight behind Every Tongue: open models perform unevenly across languages. Swahili output was meaningfully stronger, while Akuapem Twi exposed the risks of confident but weak low-resource generation. That is not a weakness we hide; it is the reason the product exists. Every Tongue combines ScriptureFlow grounding, English back-translation, uncertainty flags, and native-speaker review so ministry teams can use AI carefully instead of blindly. Commercially, Every Tongue is the first product built on the ScriptureFlow platform. It can serve individual translators for free, ministry organizations through subscriptions, and later NGOs or government teams that need content localization for underserved languages. The goal is not to replace human translators or pastors. The goal is to give them a safer AI-assisted drafting workflow for languages current AI systems often leave behind.