App Build

SENIOR REACT NATIVE DEVELOPER — AI Voice-First Cooking App (iOS & Android) We are building a voice-first AI cooking companion for iOS and Android. It is a photorealistic AI chef who guides users through cooking step-by-step, remembers their history across every session, handles mid-cook emergencies in real time, and grows a genuine relationship with the user over time. This is not a recipe app. It is a full-stack AI product with a real-time voice pipeline, structured memory architecture, a purpose-built LLM harness, and a live avatar powered by Tavus. This role is for a senior React Native developer who has shipped production mobile apps and is comfortable working across a modern AI stack. You will be the primary developer on this build. --- THE STACK • React Native (iOS + Android) • Firebase (Firestore + Authentication) • Vercel serverless functions (all API keys server-side — never in app binary) • Anthropic Claude API (recipe generation, memory, harness) • Claude Vision (fridge scan + technique analysis) • OpenAI Whisper (voice transcription) • Tavus API (photorealistic avatar, real-time lip sync, two-way camera perception) • ElevenLabs (TTS — pending Tavus voice evaluation) • DALL-E 3 (AI food photography) • RevenueCat (subscriptions) • Sentry (error monitoring) • PostHog (analytics) --- WHAT YOU WILL BUILD Core voice pipeline: wake phrase detection ("Hey Chef"), Whisper transcription, Claude response generation, ElevenLabs or Tavus TTS, all streamed with sub-1.5 second latency end to end. LLM harness: a purpose-built middleware layer around Claude that enforces persona consistency, dietary safety guardrails (allergy flags caught before output reaches the user), server-side deterministic rules (negatively rated recipes are permanently excluded), and structured memory injection on every session. Memory architecture: structured Firestore session records after every cook — dish, techniques, rescue moments, skill progression, inferred goals. A living pattern summary maintained and updated after each session, injected into every subsequent session so Chef Ai's behavior is visibly informed by the user's history over time. Tavus avatar integration: photorealistic chef avatar with real-time lip sync, idle corner loop during active steps, full activation for greetings, rescue mode, technique coaching, and sign-off. Two-way camera perception for live technique feedback (pending Tavus technical evaluation). Fridge scan: Claude Vision identifies ingredients from a camera photo and Chef Ai names a dish immediately. Rescue mode: always-on system that handles mid-cook emergencies (burned ingredients, broken sauces, missing substitutions, technique failures) by activating Chef Ai in full Tavus session mode. Session persistence: cooking session state saved continuously — app close, phone call, or interruption resumes exactly where the user left off. Hosting and entertaining mode: Chef Ai builds a complete dinner party plan through conversation — scaled recipes, guest dietary reconciliation, prep timeline, consolidated shopping list, equipment check with affiliate links, wine pairings, and push notification reminders. Subscriptions: RevenueCat integration, one free recipe before paywall, AI food photography (DALL-E 3) gated behind paid tier. Full feature list in the attached spec. --- NON-NEGOTIABLE REQUIREMENTS • 4+ years React Native in production (not Expo-only) • At least one shipped app on both App Store and Google Play • Direct experience integrating LLM APIs (OpenAI, Anthropic, or equivalent) in a production app — not just prototypes • Experience building real-time voice pipelines (STT → LLM → TTS) with latency under 2 seconds • Firebase Firestore architecture experience — you have designed data models, not just queried them • Serverless backend experience (Vercel, AWS Lambda, or equivalent) • Ability to work independently and communicate in writing, same day --- NICE TO HAVE • Prior Tavus or ElevenLabs integration • Experience with RevenueCat • Experience building a harness or middleware layer around an LLM • React Native Vision Camera experience • Fine-tuning or prompt engineering experience at production level --- ENGAGEMENT This is a full build — not a feature sprint. We expect the right developer to be the primary technical voice on this project, not an executor of tickets. We communicate in writing, respond within 24 hours, and expect the same. Target prototype delivery: October 2025. Budget and rate: open to fixed-price milestone structure, include your rate in your proposal.