Foodie Finder

Where to eat, without the endless scrolling.

Foodie Finder is a cross-platform mobile app that helps you decide where to eat. Spin a wheel of nearby restaurants, favorite the places you love, read community tips from other diners, and get personalized recommendations — including matches in other cities for when you travel.

Veteran-owned · Built by Sassy Consulting LLC

---

What it does

• 🎰 Spin the wheel. Random-pick from real restaurants near any postal code. Filter by price, cuisine, open now, and how recently you've been there.
• ⭐ Favorites are yours. Heart-mark restaurants you like. Favorites and notes live on your device only — no account, no login, no cloud sync.
• 📝 Two kinds of notes.
  • Personal notes — private, on-device, for remembering "get the fish tacos"
  • Community tips — optional public notes visible to other diners, rate-limited and content-moderated
• 🗺️ Real data from real sources. Google Places, Foursquare, HERE Maps, OpenStreetMap, and Culver's public API all merged into one result set with aggregated ratings.
• 📸 Menu photos, correctly classified. Google Vision OCR identifies which restaurant photos are actual menu pages (not food shots) and surfaces them in a dedicated menu section. You can also upload your own menu photos.
• ✨ Taste-based recommendations. Favorite a few restaurants in your home area, and the app builds a local-only taste profile. It then surfaces:
  • In Your Area — similar spots you haven't tried yet
  • When You Travel — matches in cities far from home (cross-locale matching via a favorites centroid + geographic radius filter)
• 🍦 Culver's Flavor of the Day — because Midwest.
• 🔗 Share from Google Maps. Tap "Share" on any Google Maps restaurant and Foodie Finder will import it directly.
• 🔍 Natural-language search. "Cozy Italian with outdoor seating" or "late-night food that's not fast food" — semantic search via Cloudflare Vectorize + Workers AI (production only).

Architecture

┌──────────────────────────┐
│  Expo / React Native     │  SDK 54, new arch, Hermes
│  app/                    │  iOS + Android + (partial) web
│  - (tabs)/               │  Home, Browse, Favorites, Settings
│  - restaurant/[id].tsx   │  Detail page
│  components/             │  PhotoCarousel, MenuSection, PublicNotes,
│  hooks/                  │  TasteMatches, SpinnerWheel, etc.
│  utils/                  │
└──────────┬───────────────┘
           │ tRPC over HTTPS (superjson)
           ▼
┌──────────────────────────┐
│  Cloudflare Worker       │  Hono + tRPC
│  worker/                 │  - Google Places / Foursquare / HERE / OSM scraper
│  - trpc-router.ts        │  - D1 restaurant cache
│  - scraper.ts            │  - KV for community tips + rate limit
│  - content-guard.ts      │  - R2 for menu photo uploads
│  - vector-search.ts      │  - Vectorize + Workers AI for semantic search
└──────────┬───────────────┘
           │
           ▼
┌──────────────────────────┐
│  Cloudflare edge         │
│  - D1 database           │
│  - KV (RATE_LIMIT +      │
│    FOODIE_PUBLIC_NOTES)  │
│  - R2 (MENU_PHOTOS)      │
│  - Vectorize index       │
│  - Workers AI            │
└──────────────────────────┘

Local development runs a parallel server/ tRPC on Node that mirrors the worker's routes, so you can iterate without Cloudflare.

Project structure

.
├── app/                  # Expo Router screens
├── components/           # React Native components
├── hooks/                # React hooks (local state, tRPC clients, taste profile)
├── utils/                # Pure helpers (geo, hours, PII guard, photo classifier)
├── server/               # Local-dev tRPC server (mirrors worker API)
├── worker/               # Cloudflare Worker production backend
├── types/                # Shared TypeScript types
├── constants/            # Theme, OAuth constants
├── context/, contexts/   # React context providers
├── lib/                  # Runtime bits (tRPC client, env validator)
├── scripts/              # Ad-hoc scripts (integration tests, content guard tests)
├── assets/               # Icons, fonts, splash images
├── android/              # Native Android project (committed, not managed)
├── wrangler.toml         # Cloudflare Worker config
├── app.config.ts         # Expo config
└── eas.json              # EAS Build profiles

Setup

Prerequisites

• Node 20+ (Node 24 works)
• pnpm (or npm/yarn — project is pnpm-native)
• EAS CLI: npm install -g eas-cli
• Wrangler CLI (installed via project deps)
• Android Studio + JDK 17 (for local Android builds)
• Xcode (for local iOS builds, macOS only)

Install

pnpm install

Environment variables

Copy .env.example → .env and fill in the required keys:

# Required — server-side, never committed
GOOGLE_PLACES_API_KEY=         # https://console.cloud.google.com — enable Places API
FOURSQUARE_API_KEY=            # https://foursquare.com/developers
HERE_API_KEY=                  # https://developer.here.com
JWT_SECRET=                    # any long random string

# Required — client-side (EXPO_PUBLIC_ prefix = bundled into app)
EXPO_PUBLIC_API_BASE_URL=      # dev: http://localhost:3000, prod: https://foodie-finder.sassyconsultingllc.com
EXPO_PUBLIC_GOOGLE_VISION_API_KEY=  # https://console.cloud.google.com — enable Vision API

# Optional — if you want to populate the Vectorize semantic index locally
DATABASE_URL=
CLOUDFLARE_ACCOUNT_ID=

.env is .gitignored. Never commit API keys — the project has a linter check for that.

Run locally

# Start the Metro bundler + the local tRPC server together
pnpm dev

# Or run them separately:
pnpm dev:metro       # Expo Metro bundler
pnpm dev:server      # tsx watch server/_core/index.ts

# Worker (Cloudflare local dev)
pnpm worker:dev

Scan the QR code with Expo Go or your dev client, or press a / i to open the Android / iOS simulator.

Testing

# Unit + integration tests (vitest)
pnpm test

# Typecheck the whole project
npx tsc --noEmit

# Lint + fix
pnpm lint

# Expo health check
npx expo-doctor

# Integration test: real Google Places data + cross-locale taste matching
pnpm tsx scripts/test-cross-locale-matching.ts

# Content guard unit test
pnpm tsx scripts/test-content-guard.ts

Deployment

Mobile (Expo + EAS Build)

# Preview APK for internal testing
npx eas-cli build --platform android --profile preview

# Production AAB for Play Store
npx eas-cli build --platform android --profile production

# iOS builds (requires Apple Developer account)
npx eas-cli build --platform ios --profile preview

Build profiles live in eas.json. The project uses the committed android/ native folder (not managed prebuild), so app.config.ts fields like plugins, ios, android are informational — the native files are the source of truth.

Backend (Cloudflare Worker)

# Deploy to preview environment
npx wrangler deploy --env preview

# Deploy to production
npx wrangler deploy --env production

The worker is configured in wrangler.toml. It binds:

• DB → D1 (foodie-finder)
• FOODIE_PUBLIC_NOTES → KV (community tips)
• RATE_LIMIT → KV (per-IP rate limiter)
• MENU_PHOTOS → R2 (user menu photo uploads)
• VECTORIZE → Vectorize index (semantic search)
• AI → Workers AI (embeddings)

API keys (GOOGLE_PLACES_API_KEY, FOURSQUARE_API_KEY, HERE_API_KEY, JWT_SECRET) are set via wrangler secret put, NOT in wrangler.toml.

Feature deep-dives

Menu photo classification

hooks/use-classified-photos.ts → utils/photo-classifier.ts runs Google Cloud Vision TEXT_DETECTION on Google Places photos, treats photos with dense text (≥250 chars, ≥40 words, ≥8 lines) as menu pages, and routes them to the MenuSection component. Results are cached per-URL in AsyncStorage so re-opening a restaurant is free. Food photos go to the PhotoCarousel; menu photos go to the MenuSection (capped at 5).

Taste profile + cross-locale matching

hooks/use-taste-profile.ts builds a profile from the user's favorites:

• Cuisine weights — normalized counts, excluding generic tokens like "Restaurant" / "Food" / "establishment"
• Price weights — $/$$/$$$/$$$$ distribution
• Rating centroid — mean aggregated rating
• Cuisine coverage — fraction of favorites with a specific (non-generic) cuisine tag
• Favorites centroid — geographic center (lat/lng) of all favorites

Scoring is adaptive: if cuisine coverage is high, cuisine dominates. If coverage is low (user favorited mostly "Restaurant" genericos), the scorer falls back to price + rating proximity. Cross-locale filtering uses haversine distance from the favorites centroid, NOT city-name string matching, so adjacent small towns are correctly treated as "local" rather than "travel."

Community tips + content guard

components/public-notes-section.tsx posts tips via the addPublicNote tRPC procedure. Moderation happens on BOTH the client (utils/pii-guard.ts, advisory) and the worker (worker/content-guard.ts, enforcing). The worker:

1. Rate-limits per cf-connecting-ip via the RATE_LIMIT KV namespace (10 posts/hour)
2. Runs guardPublicNote() — hard-blocks profanity, slurs, threats, drug promotion, and narrowly-scoped staff harassment ("fire the manager")
3. Scrubs PII (phone, email, SSN, credit card) and replaces with [type removed] placeholders
4. Dedupes identical tips from the same IP within 60s
5. Caps per-restaurant notes at 200 (rotating oldest out)

The patterns are intentionally narrow so legitimate complaints like "food sucks" or mentions of "fire sauce" are NOT blocked. 13/13 test cases pass in scripts/test-content-guard.ts.

Security posture

• No user accounts — no password storage, no JWT refresh tokens, no OAuth
• Favorites / notes / preferences are device-local only
• API keys loaded exclusively from environment variables (server/_core/env.ts::ENV)
• Community tip moderation enforced server-side with narrowly-scoped regex rules
• PII auto-scrubbed from user-generated content
• Per-IP rate limiting on write endpoints
• HTTPS/TLS everywhere
• .env, keys/, *.pem, *.p12, *.jks all excluded from both .gitignore and .easignore

See COMPLIANT_PRIVACY_POLICY.md for the full privacy policy and TERMS_OF_SERVICE.md for terms.

Contributing

This is a private repository. If you've been given access and want to contribute:

1. Branch from master
2. Run npx tsc --noEmit && pnpm test before pushing
3. Keep commits atomic and descriptive
4. No secrets in source — use .env
5. New features that add data flows require a privacy policy update

License

Proprietary. © 2025 Sassy Consulting LLC. All rights reserved.

Contact

Email: info@sassyconsultingllc.com Website: https://sassyconsultingllc.com Privacy Policy: https://privacy.sassyconsultingllc.com/foodie-finder

---

Built with ♠ by Sassy Consulting LLC — a Veteran-Owned company.