Case studies — AI APP
baba
Context-aware Hebrew translator, not a dictionary lookup, but a system that understands register, gender, and formality.
ZipLyne designed and shipped a full cross-platform product: web app, iOS and Android apps, and a Chrome extension that injects into any webpage. Under the hood, 11 gender/formality prompt variants handle Hebrew's grammatical complexity through an OpenRouter multi-model backend on Cloudflare Workers, with ElevenLabs TTS for native-speaker pronunciation on every result.
Four platforms. One system.
One codebase on Expo spanning iOS (App Store) and Android (Google Play). Separate React web app. Chrome extension that translates selected text in-place without leaving the page.
Masculine/feminine × formal/informal combinations, each with tailored system prompts. Context detection routes every query to the correct variant before hitting the LLM, no post-hoc correction.
Worker handles routing, rate-limiting, and model selection. OpenRouter provides multi-model fallback, if the primary model rate-limits, traffic fails over automatically. Zero cold starts.
Every translation surfaces a playback button. ElevenLabs streams native-quality Hebrew audio on demand, no pre-recorded files, no static dictionary clips.
The AI chain.
User input parsed for gender cues, register signals, and explicit formal/informal preference.
One of 11 gender/formality prompt templates selected and injected with the query.
OpenRouter routes to primary model with automatic fallback. Returns Hebrew + transliteration + usage notes.
Hebrew output piped to ElevenLabs for native pronunciation audio, streamed to client on demand.
What it runs on.
On the bench.
- Cloudflare Workers
- OpenRouter
- ElevenLabs
- React
- Expo
- Chrome Extension APIs