Panko - AI Fitness Coach App

An AI fitness coach for iPhone and Android that logs meals from text, voice or a photo, tracks gym sessions and personal records, and reviews your week. Built with Expo and React Native on Firebase, with a Gemini-powered coach that can read your training history and take actions in the app. Live on Google Play and the App Store as a subscription product.

Overview

Panko is a pocket AI fitness coach built under my product studio PankoByte. Most fitness apps hand you a database and leave the thinking to you; Panko brings food logging, workout tracking and coaching into one app and does the number-crunching for you. It is available on Google Play and the App Store with monthly and yearly subscriptions and a 7-day free trial, and has real users training with it.

Key Features

Log food in seconds: describe a meal ("chicken rice bowl and a banana"), snap a photo or say it out loud; Panko estimates portions, calories and macros and files it under the right meal
A–E food grades: every logged food gets a European-style nutrition grade with a short reason
Workout tracking: last session's weight and reps sit above each set, the rest timer runs itself, and personal records are caught the moment you hit them
AI coach chat: ask by text or voice ("I have twenty minutes and no barbell"); the coach reads your sessions and meals before answering and can build and save a workout for you
Weekly review: training, food quality, nutrition targets and lift progress for the past week, with a personal note from Coach Panko
Progress: strength charts per lift, a smoothed weight trend, streaks and a consistency calendar
High-protein recipes filtered by diet and meal
1,300+ exercises with form notes, demos and swaps for the equipment your gym has
Home-screen widgets for calories, streak and the next session, plus opt-in reminders
Private progress photos that never leave the phone

AI Coach Architecture

The coach runs on Gemini through Firebase AI Logic, with a tool layer (src/data/agent) that turns it from a chatbot into an agent:

Tools let the model search the exercise catalog, read past sessions and nutrition, log food and create workouts on the user's behalf
Context-aware answers: before replying, the coach reads the sessions the user actually did, so advice is grounded in real history rather than generic tips
Multimodal logging: text, voice and photo meal entries are parsed into structured food items with macros and a grade
Usage metering: AI usage is counted server-side in Firestore (aiUsage/{uid}) and enforced by security rules as a guard against runaway loops

Technical Implementation

Mobile App:

Expo SDK 56 + React Native 0.85 + TypeScript, file-based routing with Expo Router
Reanimated, Gesture Handler and Vision Camera for a fluid, native-feeling UI
Stale-while-revalidate data layer (useCachedResource) so screens render from cache before the network answers
The 1,300+ exercise catalog ships inside the binary with pre-rendered poster frames, generated from Firestore by build scripts
Native Android widgets (react-native-android-widget) and iOS widget targets, plus local Expo modules and config plugins
Progress photos stored only in app storage with Android auto-backup disabled

Backend:

Firebase: Auth (Google & Apple sign-in), Firestore, Storage, App Check and Crashlytics
Cloud Functions for scheduled sweeps (weekly recaps, streak-risk nudges, trial-ending and inactivity emails), a transactional email outbox and the RevenueCat webhook
RevenueCat for subscriptions across Google Play and the App Store, with a single entitlement as the paywall gate

Tooling & Ops:

EAS Build & Submit release pipeline
A local Next.js admin console for reading and replying to in-app feedback via the Admin SDK
A prerendered marketing site with a blog and free fitness calculators (calorie, macro, BMI, 1RM, water intake)

Demo Video

Screenshots