Purr

Overview

Purr is a mobile app for practicing hard conversations before you have them. Pick a scenario (a job interview, a first date, a difficult talk with a manager) or write your own, then rehearse it out loud against an AI character that plays the other side in real time. When you finish, you get scored and told what to fix. I co-built it, and it has passed 1,000 downloads across the App Store and Google Play.

What I built

• Real-time voice. Audio streams over WebRTC to a FastAPI server running Deepgram for transcription and the OpenAI Realtime API for the character. Latency had to feel like a phone call, not a chatbot.
• A parallel evaluation pipeline. After each session, independent judges score content and delivery separately, 0 to 5 stars, so the feedback is specific instead of "nice job".
• The app itself. React Native with Expo, curated lesson suites, custom scenarios, progress tracking, and Apple, Google, and email sign-in.

What I learned

This was my first real-time system and my first app in a store. Shipping something strangers download and rate is a different kind of accountability than shipping to a team.