Skip to content

README.md

Latest commit

 

History

History
186 lines (150 loc) · 10.7 KB

README.md

File metadata and controls

186 lines (150 loc) · 10.7 KB
Raw

Reveil — AI-Powered Adaptive Habit Transformation

Graduation project (COMP4901, Konya Food and Agriculture University). Reveil is a mobile-first habit formation and addiction recovery app that combines behavioral science with adaptive, AI-generated feedback. The implementation tracks the project's SRS and SDD.

Layer Stack Path
Mobile Expo SDK 54, React Native 0.81, TypeScript, Firebase Auth mobile/
Backend NestJS 11, TypeORM, SQLite (dev) / Postgres (prod), Firebase Admin backend/
AI Engine FastAPI, OpenAI (optional), rule-based fallback, safety filter ai_engine/
Infra docker-compose (Postgres) docker-compose.yml

Architecture at a glance

┌────────────┐     HTTPS / JWT      ┌────────────┐     HTTP      ┌────────────┐
│ Mobile App │ ───────────────────► │  Backend   │ ────────────► │ AI Engine  │
│  (Expo)    │ ◄─────────────────── │  (NestJS)  │ ◄──────────── │ (FastAPI)  │
└─────┬──────┘                      └─────┬──────┘               └─────┬──────┘
      │                                   │                            │
      │ Firebase Auth                     │ TypeORM                    │ OpenAI (opt.)
      ▼                                   ▼                            ▼
   Firebase                            SQLite/Postgres                OpenAI API

See docs/architecture.md for view-by-view diagrams, docs/TRACEABILITY.md for the SRS-to-code mapping, and docs/DEFENCE_NOTES.md for the project-defence cheat sheet.

Feature highlights

  • Habits with categories, weekly targets (1–7×/week), pause-until auto-resume, freeze days that protect streaks, sort, search, 8 quick-start templates including 4 psychoactive-cessation templates (smoking, alcohol, social media, caffeine), and per-habit accent color (8 presets, NULL → brand fallback).
  • Tracking with mood (1–10) and notes; idempotent same-day upsert.
  • AI feedback with locale (TR/EN), category-aware hints, and tone selector (coach / gentle / firm / playful). Provider chain Claude (ANTHROPIC_API_KEY) → OpenAI (OPENAI_API_KEY) → Groq (GROQ_API_KEY) → deterministic rule-based fallback. Override order via LLM_PROVIDER_ORDER. Recovery-category habits get a withdrawal-aware milestone narrative (day 1 acute → day 3 craving peak → day 7 high-relapse window → day 14 dopamine recalibration → day 30 identity). All output passes a safety filter that blocks substance-encouraging phrases (just one drink, bir kadeh zarar vermez, etc.).
  • Recovery support — every habit in the recovery category surfaces a one-tap crisis card linking to YEDAM 115, SABİM 184, 112, 988, Samaritans 116 123, plus a built-in 4-7-8 breathing exercise screen (clinically validated, 76 s for 4 cycles) reachable from the crisis Alert before any phone call.
  • Dashboard with consistency, streak, longest streak, weekly bar chart, GitHub-style 90-day heatmap (tinted to each habit's accent color), per-category rollup, badge strip with celebration toast, weekly insight card, mini per-habit sparklines, XP/level bar, streak milestone toasts at 7 / 14 / 30 days (recovery-aware text variants), and a Mood × Consistency Pearson correlation insight aggregated across all habits.
  • Reminders (cron-driven push) with quiet hours and per-user opt-out for the weekly digest.
  • Notification feed persisting every push the system sends.
  • Privacy — JSON + CSV export, JSON + CSV import, account deletion, opt-out toggles.
  • Mobile polish — onboarding with AI tone quiz (saves default to AsyncStorage), animated splash loader (☀️ Reveil mark + fade-in slide-up), Light / Dark / System theme with live OS color-scheme tracking, compact density mode, TR/EN i18n, accessibility (roles, states, hints), offline banner, haptics on log, success toast on Mark Done + freeze toast, level-up + badge unlock toasts, share streak text + share progress card SVG, skeleton loaders on Dashboard while loading, empty-state CTAs across 4 screens, pull-to-refresh on Reminders / Feedback / Notifications, long-press to copy AI feedback to clipboard, search-result highlighting on habits list, in-app glossary (6 behavioral-science terms with ⓘ tooltips), 3-step app tour shown once on first launch.
  • Operability — Sentry (no-op when DSN missing) on both ends, /api/health (now including AI engine round-trip latency + status), /api/version (now exposing git sha + build time), @nestjs/throttler rate limits with stricter cap on AI feedback (user-facing 429 banner via global toast), helmet headers, audit log of 5xx errors at /api/admin/errors, platform insights at /api/admin/insights.

Quick start

Prerequisites: Node 20+, npm, Python 3.11+, Expo CLI (npx expo), optional Docker.

# 1. Backend (defaults to SQLite, mock auth)
cd backend
npm install
npm run start:dev

# 2. AI engine
cd ../ai_engine
python -m pip install -r requirements.txt
python -m uvicorn main:app --port 8000

# 3. Mobile
cd ../mobile
npm install
npm start          # press 'a' for Android, 'i' for iOS, 'w' for web

API is at http://localhost:3000/api, Swagger UI at /api/docs, OpenAPI JSON at /api/docs-json. AI engine listens on http://localhost:8000.

Seed demo data

cd backend
npm run seed       # creates a demo user, 3 habits, 30 days of synthetic logs

Configuration

Backend (.env.example):

Variable Purpose Default
PORT HTTP port 3000
AI_ENGINE_URL AI engine base URL http://127.0.0.1:8000
DB_DRIVER sqlite or postgres sqlite
DB_HOST / DB_PORT / DB_USER / DB_PASSWORD / DB_NAME Postgres only localhost / 5432 / postgres / postgres / reveil
DB_PATH SQLite file path reveil.sqlite
DB_SYNC TypeORM auto-sync (dev only); set false and use npm run migration:run in prod true
FIREBASE_SERVICE_ACCOUNT Inline JSON for Firebase Admin SDK empty → mock auth
GOOGLE_APPLICATION_CREDENTIALS Alternative path-based credentials empty
FIREBASE_PROJECT_ID Project ID when using ADC empty
ADMIN_TOKEN / ADMIN_UID Gate /api/admin/* endpoints empty
DIGEST_CRON Set to off to disable the Sunday weekly-digest push on
RESEND_API_KEY / DIGEST_FROM_EMAIL Optional email digests via Resend empty → push only
SENTRY_DSN / SENTRY_TRACES_SAMPLE_RATE Optional error reporting empty → no-op

AI engine (.env.example):

Variable Purpose
AI_PROVIDER Force a single provider: claude, openai, or rule. Empty → auto (Claude → OpenAI → Rule)
ANTHROPIC_API_KEY Enables Claude path
ANTHROPIC_MODEL Claude model name (default claude-haiku-4-5)
OPENAI_API_KEY Enables OpenAI path
OPENAI_MODEL OpenAI model name (default gpt-4o-mini)

Mobile (EXPO_PUBLIC_* env so the values reach JS at runtime):

Variable Purpose
EXPO_PUBLIC_API_URL Override backend URL (defaults to http://10.0.2.2:3000/api on Android, http://localhost:3000/api elsewhere)
EXPO_PUBLIC_FIREBASE_API_KEY / _AUTH_DOMAIN / _PROJECT_ID / _APP_ID Firebase web config; absent → mock auth
EXPO_PUBLIC_GOOGLE_CLIENT_ID_IOS / _ANDROID / _WEB Enable Google sign-in via expo-auth-session
EXPO_PUBLIC_SENTRY_DSN Mobile error reporting (no-op when empty)

Postgres via docker-compose:

docker compose up -d postgres
DB_DRIVER=postgres npm run start:dev      # in backend/

Production migrations:

DB_DRIVER=postgres DB_SYNC=false npm run migration:run

Testing

# Backend unit + e2e
cd backend && npm test && npm run test:e2e

# AI engine
cd ai_engine && python -m pytest

# Mobile typecheck + jest
cd mobile && npx tsc --noEmit && npm test

GitHub Actions (.github/workflows/ci.yml) runs all four suites plus a Postgres-backed e2e job that verifies migrations idempotently.

Project layout

backend/
  src/
    auth/                  Firebase guard + admin SDK wrapper
    users/                 Entity, preferences, JSON/CSV export & import
    habits/                CRUD, reorder, search, pause-until
    tracking/              Daily completion logs
    feedback/              AI feedback proxy + share-card SVG
    dashboard/             Aggregated metrics + category rollup + level
    notifications/         Push tokens, reminders, scheduler, feed, events
    digest/                Weekly digest service + Sunday cron + email
    insights/              Admin platform-wide stats
    audit/                 5xx error audit log + Sentry forwarder
    health/                /health and /version endpoints
    common/                Stats / badges / level helpers
    migrations/            Driver-aware initial schema
  test/                    Supertest e2e
ai_engine/
  main.py                  FastAPI app
  feedback.py              Prompt builder, safety filter, rule fallback
  test_feedback.py         pytest
mobile/
  src/
    components/            UI primitives + Heatmap, Sparkline, ProgressRing,
                           XpBar, BadgeStrip, BadgeUnlockToast, LevelUpToast,
                           CategoryBreakdown, MoodInsight, MoodPicker, Toggle
    navigation/            Stack navigator (with deep-link ref)
    screens/               Login, Onboarding, Dashboard, Habits, CreateHabit,
                           HabitDetail, Feedback, Reminders, Digest,
                           NotificationFeed, Profile
    services/              api, auth, firebase, social-auth, sentry,
                           notifications, share, haptics, density, i18n,
                           preferences
    theme/                 Dark theme tokens
docs/
  architecture.md          Mermaid diagrams of all five SDD viewpoints
  TRACEABILITY.md          SRS REQ-1..REQ-8 mapped to code
  DEFENCE_NOTES.md         One-page jury cheat sheet
scripts/
  generate_brand_assets.py Pillow-based icon/splash/SVG mark generator

Authors

  • Halil Utku DEMİRTAŞ (222010020054)
  • Furkan Can KARAFİL (222010020013)

Supervisor: Prof. Dr. Meltem Huri BATURAY KHAN.