The Aura Orb breathes, shifts colour with risk, and shows your burnout index at a glance.
India's 300 million students and young professionals face a silent epidemic: burnout. Unlike sudden illness, burnout builds gradually over days and weeks — yet no accessible tool warns users before it strikes.
Existing mental health apps rely on static questionnaires, require expensive subscriptions, or store sensitive data on remote servers.
There is no voice-first, privacy-preserving, AI-powered companion that can detect early burnout signals, learn from daily emotional patterns, and proactively intervene — specifically designed for India's students and early-career workforce.
AuraPulse is a voice-first, on-device Android wellness companion.
Each day, a 2-minute natural conversation with an AI (powered by Agora ConvoAI) captures the user's emotional state — no forms, no typing. Gemini then analyses the transcript to extract mood, energy, and stress signals.
Over 7–14 days, a local prediction engine identifies dangerous burnout trajectories and delivers timely, personalised nudges — breathing exercises, social prompts, rest reminders — before the user hits crisis.
All data stays on the device. No account needed. Free forever.
Home — dark theme Aura Orb showing the on-device burnout index, with the live AI Coach card right below. |
Home — light theme Every surface adapts cleanly via mirrored DayNight tokens. |
Quick actions The 5-emoji quick check-in and the premium Talk to your AI Coach voice CTA — the marquee feature. |
Chat with AuraPulse Multi-turn typed conversation. Stays warm even when the API quota is dead — keyword-aware empathetic fallback. |
Trends Burnout gauge, at-a-glance insight pills, and a 14-day mood line. |
AI Weekly Digest 7-day energy bars, detected patterns and a 3-day forecast — computed on-device, no API call. |
History Every past check-in with a colour-coded dominant emotion — calm=blue, tired=amber, overwhelmed=red, content=green. |
Breathing nudge Auto-suggested on HIGH-risk days. Pick a pattern, follow the breathing circle. |
- 2-minute voice check-in — real spoken conversation via Agora ConvoAI + Cartesia STT/TTS. No forms, no typing.
- Typed Chat with AuraPulse — voice fallback that still feels warm during free-tier quota exhaustion thanks to a keyword-aware scripted mode.
- Quick check-in (QnA) — 5 sliders, deterministic scoring, ~30 seconds. The fast path on busy days.
- AI Coach card on Home — live Gemini-generated nudge conditioned on your last 7-day mood trend. Cached 6 hours; refresh on demand.
- "Why this score?" explanation — after every check-in, Gemini explains in plain language which signals drove your mood and energy numbers.
- On-device burnout engine — pure-Kotlin weighted scoring over 14 days of mood + 7 days of energy, producing a 0-100 score with LOW / MODERATE / HIGH bands.
- Trends + AI Weekly Digest — line + bar charts, insight pills, detected patterns, 3-day mood forecast — all rules-based and on-device.
- Colour-coded emotion vocabulary — stressed=red, tired=amber, calm=blue, happy=green, hopeful=violet. History reads at a glance.
- Smart nudges — when risk crosses 65, a 4-7-8 breathing exercise is suggested automatically.
- Privacy by construction — audio is ephemeral and never persisted; only text crosses the network; no account, no cloud sync.
- DayNight theming — mirrored light + dark colour tokens. Every card, chart and drawable adapts.
flowchart TD
A([All data starts as user words — voice or typed])
A --> V[VOICE PIPELINE<br/>Agora SDK → Cartesia STT<br/>ConvoAI Agent with Gemini brain<br/>on Render backend]
A --> C[CHAT PIPELINE<br/>Multi-turn Gemini<br/>Keyword fallback on quota errors]
A --> Q[QnA PIPELINE<br/>Deterministic 5-slider scoring<br/>mood and energy]
V --> G{Gemini 2.0 Flash<br/>JSON: moodScore, energyLevel,<br/>dominantEmotion, stressIndicators, summary}
C --> G
Q --> G
G -->|Gemini reachable| P[Structured JSON<br/>parsed and clamped to 1..10]
G -->|Rate-limited or offline| L[Local keyword analyzer<br/>'stressed' → mood 3<br/>'amazing' → mood 8]
P --> M[(MoodEntry persisted<br/>Room — one row per day)]
L --> M
M --> E[Burnout Engine<br/>pure Kotlin, on-device<br/>14-day weighted window]
E --> R{{Burnout Risk 0-100<br/>LOW / MODERATE / HIGH}}
R --> O[Aura Orb on Home<br/>color-shifting breathing orb]
R --> K[AI Coach card<br/>Gemini-generated personalised advice<br/>cached 6 hours]
R --> N[Smart nudges<br/>4-7-8 breathing, social prompt,<br/>rest reminder]
classDef pipeline fill:#1f2a4a,stroke:#5EA8FF,color:#fff,stroke-width:2px
classDef ai fill:#2d2150,stroke:#A78BFA,color:#fff,stroke-width:2px
classDef store fill:#0F2C3A,stroke:#7DD3FC,color:#fff,stroke-width:2px
classDef output fill:#1A3A50,stroke:#4ADE80,color:#fff,stroke-width:2px
class V,C,Q pipeline
class G,P,L,E ai
class M store
class R,O,K,N output
Free-tier AI quotas are unforgiving. We engineered around it.
| Failure mode | Defence |
|---|---|
| Google deprecates a model | Auto-fallback chain: gemini-2.0-flash → gemini-flash-latest → gemini-1.5-flash-latest → gemini-2.5-flash |
| 429 rate-limit | One transparent retry after 800 ms; 60-second rate-lock if it persists |
| AI Coach burning calls | 6-hour in-memory cache + 10-second tap debounce — ≤ 4 calls/day per user |
| Chat hitting quota | 10-minute scripted-mode lock with keyword-aware empathetic replies — feels like real chat, not a stock fallback |
| Transcript → analysis offline | Local keyword analyser — "stressed" → mood 3, "amazing" → mood 8. Honest, never neutral by default. |
| Render backend cold-starts | App pings /ping on launch + UptimeRobot keeps it warm |
| Self-harm mentions in chat/voice | Hard-coded helpline routing (iCALL, Vandrevala) fires before any API call |
| Stale Room row on re-check-in | deleteByDate(today) runs before insert — newest check-in always wins |
| Layer | Choice |
|---|---|
| Language | Kotlin (100%, no Java) |
| UI | XML layouts + ViewBinding (no Jetpack Compose by design — native, fast, IDE-preview friendly) |
| Architecture | Clean Architecture — presentation/ → domain/ → data/ |
| Pattern | MVVM with StateFlow<UiState<T>> |
| DI | Hilt |
| Async | Kotlin Coroutines + Flow + StateFlow |
| Persistence | Room (3 tables) + DataStore (preferences) |
| Networking | Retrofit + OkHttp + Gson |
| Charts | MPAndroidChart |
| Background | WorkManager (daily reminder + morning recompute) |
| Voice | Agora SDK + Agora ConvoAI + Cartesia STT/TTS |
| LLM | Gemini 2.0 Flash via REST |
| Backend | Flask + gunicorn on Render (tools/agent_server.py) |
| Theming | DayNight + custom token system (light + dark mirrored) |
| Screen | What it does |
|---|---|
| Home | Aura Orb, AI Coach card, emoji quick check-in, premium voice CTA, chat / QnA / breathing entries |
| Voice check-in | Real spoken conversation via Agora ConvoAI; live transcript; auto-ends or you tap Finish |
| Chat with AuraPulse | Typed multi-turn Gemini conversation; scripted fallback; tap Finish to analyse |
| Quick check-in | 5-slider deterministic scoring, ~30 sec |
| Analysis result | Mood / energy / colour-coded emotion / summary + "Why this score?" Gemini explanation |
| Trends | Burnout gauge, insight pills, 14-day mood line, 7-day energy bars, AI Weekly Digest with forecast |
| History | All past check-ins with colour-coded emotions; tap a row for the full detail |
| History detail | Mood / energy / emotion / stress indicators / summary / "What you said" transcript |
| Breathing nudge | 4-7-8 / Steady / Balance breathing exercise — animated circle |
| Onboarding | Name → age group → primary stress source (one-time, personalises prompts) |
| Profile / Settings | Theme toggle, name edit, "Erase my data" wipe-on-device |
| Guarantee | Enforcement |
|---|---|
| Audio never persisted | Agora SDK streams in-memory only; no file written on device or backend |
| Only text crosses the network | The transcript is the only thing Gemini receives — no audio, no identifiers |
| No account, no login, no sync | All state lives in local Room + DataStore. No user table on any server. |
| Burnout score is local-only | Computed by ComputeBurnoutRiskUseCase on-device — Gemini never sees it |
| Self-harm safety net is offline-first | Keyword guard routes to iCALL / Vandrevala helplines before any API call |
| One-tap data wipe | Settings → "Erase my data" drops the Room database |
- Android Studio Ladybug or newer (
compileSdk 35) - A physical device or emulator on Android 7.0+ (API 24+)
- Internet for the AI surfaces (Gemini key in
local.properties)
git clone https://github.com/Phoenix1808/AuraPulse.git
cd AuraPulse
copy local.properties.example local.properties
notepad local.properties # fill in GEMINI_API_KEY at minimum
# build + install on a connected device
.\gradlew clean assembleDebug installDebug# REQUIRED — powers AI Coach, Chat, mood analysis, "Why this score?"
# Get a free key at https://aistudio.google.com/apikey
GEMINI_API_KEY=ya29...
# OPTIONAL — needed only for the voice check-in surface
AGORA_APP_ID=...
AGORA_TEMP_TOKEN=...
# OPTIONAL — URL of the Flask backend (tools/agent_server.py) on Render
AGENT_SERVER_URL=https://your-render-instance.onrender.comSecrets are read at build time and surfaced via BuildConfig — they never appear as plaintext strings in the APK.
tools/agent_server.py is a small Flask app that mints fresh Agora RTC tokens per session and spawns the ConvoAI agent (Cartesia TTS + Gemini LLM). Deploy with the included render.yaml, or run locally:
cd tools
cp .env.example .env
pip install -r requirements.txt
python agent_server.pyShipped
- Voice check-in via Agora ConvoAI + Cartesia + Gemini
- Typed Chat with AuraPulse (voice fallback, scripted-mode resilience)
- Quick check-in QnA with transparent slider scoring
- AI Coach card on Home (live Gemini, 6h cache, offline-friendly fallback)
- On-device burnout engine + Aura Orb + risk bands
- Trends — gauge, charts, insight pills, AI Weekly Digest
- "Why this score?" Gemini explanation on result screen
- Colour-coded dominant emotion in History / Detail / Result
- DayNight theming, mirrored light + dark tokens
- Self-harm safety net (iCALL, Vandrevala)
- Streak tracking + milestone celebration
Planned
- Voice-fallback banner inside the voice screen ("Voice not catching? → Chat")
- Wear OS heart-rate as an additional signal
- Localisation (Hindi, Tamil, Telugu, Bengali)
- Family / group wellness mode (opt-in shared trends)