This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

I built EchoTrail: The Invisible Bird Guide—a completely offline-first, on-device AI bird sound identifier and observation companion designed specifically to get you off the screen and into the woods.

Birdwatching is one of the most rewarding ways to disconnect from daily digital noise and immerse oneself in nature. But for beginners standing on a wooded trail, hearing an unfamiliar call in the canopy often leads to frustration:

• You hear it, but you don't know what it is or where to look.
• Backcountry connectivity is nonexistent: Most commercial identification apps stream audio to cloud servers. The moment you step into a state park, dense forest, or mountain trail, they fail completely.
• Screen fixation defeats the point of being outside: Many nature apps treat wildlife like a gaming checklist, keeping your face glued to the glass.

EchoTrail flips this paradigm with a 5-step nature discovery workflow:

flowchart LR
    A["🌲 Listen\n5–10s Trail Recording"] --> B["⚡ Identify\nOn-Device Local AI"]
    B --> C["🎧 Compare\nNative Reference Calls"]
    C --> D["👁️ Spot\nActionable Canopy Clues"]
    D --> E["📓 Discover\nPrivate Field Journal"]

1. Listen: One tap records 5–10 seconds of ambient birdsong directly on the trail.
2. Identify Locally: An on-device acoustic classifier analyzes spectral and harmonic profiles locally in under a second—completely offline with zero internet required.
3. Compare with Reference Calls: Listen to authentic audio reference calls for each candidate directly on-device to confirm subtle pitch and timbre differences with your own ears.
4. Actionable Spotting Clues: EchoTrail provides practical field observation guidance—where in the canopy the species perches, peak activity hours, and diagnostic behavioral habits.
5. Private Field Journal & Challenges: Save observations to a private local SQLite database with hearing vs. spotted confirmation, and complete outdoor challenges (First Song Identified, Canopy Explorer, Morning Chorus).

Demo

• Direct Standalone APK Download: Download echoTrail.apk
• EAS Build Dashboard & QR Code: Expo Build 4b6670a1

Code

The entire project is open-source under the MIT license:

npm run reset-project

• GitHub Repository: https://github.com/Dev-Saurabh-K/echoTrail

How I Built It

EchoTrail is built with React Native, Expo SDK 57, TypeScript, and NativeWind:

graph TD
    subgraph UI ["User Interface (Expo Router & NativeWind)"]
        Record["Record Screen (AudioWaveform)"]
        Results["Results Screen (CandidateCard & ConfidenceBadge)"]
        Species["Species Detail (ReferenceAudioPlayer & Clues)"]
        Journal["Field Journal (ObservationCard)"]
        Challenges["Challenge Dashboard"]
    end

    subgraph Core ["Local Core Engine"]
        Audio["Audio Capture & Quality Validation (SNR/Duration)"]
        Classifier["On-Device Bird Classifier"]
        DB["Local Database (expo-sqlite)"]
        Engine["Idempotent Challenge Engine"]
    end

    Record --> Audio
    Audio --> Classifier
    Classifier --> Results
    Results --> Species
    Results --> DB
    DB --> Journal
    DB --> Engine
    Engine --> Challenges

1. Robust Acoustic Validation & Uncertainty Communication

Real outdoor environments are full of wind buffeting, rustling foliage, and ambient river sounds. EchoTrail incorporates strict input validation:

• Duration Checks: Enforces a minimum 2.0s recording window to ensure sufficient acoustic evidence.
• Silence & SNR Thresholding: Rejects inaudible or below-threshold (-50dB SNR) recordings with actionable guidance instead of hallucinating.
• Transparent Uncertainty: When noise is high or scores are close, EchoTrail transparently labels predictions as "Uncertain" with explanatory notes rather than guessing.

2. Embedded Database & Privacy

Observations are stored in local SQLite tables (observations, challenges, species_metadata) using expo-sqlite. Users have complete control to review, export, or permanently erase all recorded audio and observation history directly from the settings tab.

3. Automated Verification Suite

We engineered an automated test suite (npm test) validating:

• Species catalog integrity and attribution
• Audio duration and silence rejection
• Prediction ranking and candidate formatting
• Uncertainty communication under adverse acoustic conditions
• Idempotent challenge progression and duplicate award prevention

Why Does Open Innovation Matter?

For a project built for trails and backcountry exploration, open-source, on-device AI wasn't just an implementation choice—it was the only architecture that made sense:

• True Backcountry Independence: Real trails don't have cell towers. Closed-source APIs that require streaming uncompressed audio to cloud servers break the moment you enter deep woods. By running inference locally on-device, EchoTrail works anywhere on Earth—from remote national parks to deep mountain canyons.
• Strict Location & Audio Privacy: Nature observation logs can reveal sensitive habitats and personal movement patterns. With EchoTrail, no audio recordings, location coordinates, or device telemetry ever leave your device.
• Screen Minimization: Commercial apps optimize for ad views and engagement time. Open innovation allowed us to build an app designed to do the exact opposite: make the phone the shortest part of the experience so your eyes stay on the canopy.
• Zero Ongoing Infrastructure Costs: Because inference runs on the user's phone, the app costs zero dollars in ongoing API tokens or cloud GPU servers to operate.

My Agent Session

We leveraged Antigravity and DevRelay during our hacking session to architect the on-device acoustic pipeline, package the Android APK on EAS, and verify test suites:

npx eas-cli build -p android --profile preview

Build finished successfully: https://expo.dev/artifacts/eas/VDI8UsbhoH4rqVw1aV301HPV9fOnDO5NQU74smfohTg.apk

Field Test Reflection

Taking EchoTrail outside into the morning air transformed a routine walk into an interactive field safari. Hearing a crisp, whistling series of notes in an oak tree, recording a 6-second sample, and having the on-device engine instantly rank the American Robin with reference audio confirmed the call without ever having to unlock cell data. Best of all, reading the spotting clues ("Listen for robin cheer-up, cheerily song; look at ground level in open lawns or mid-canopy perches") allowed us to look up, locate the bird with binoculars, and log a confirmed spotting into our journal.

That is the essence of Touch Grass—technology stepping out of the way so nature can take center stage.