Skip to content

HarishK21/RespiraSnap

Repository files navigation

RespiraSnap 🫁

Built for Hack Canada 2026 🇨🇦

Watch the demo here: https://youtu.be/8gZiano6k90

RespiraSnap is an AI-powered respiratory analysis tool that provides immediate, actionable insights into your breathing patterns. By capturing a single 15-second breathing snapshot, the platform analyzes audio waveforms to detect inhales, holds, exhales, and irregularities, forming a comprehensive view of your current respiratory state.

In today's fast-paced world, breathing mechanics are often overlooked, yet they are a fundamental indicator of stress, focus, and overall nervous system regulation. RespiraSnap bridges the gap between subjective feeling and objective data, turning the microphone you already have into a powerful biosensor.

image

How It Works

  1. Capture: The user records 15 seconds of live breathing audio (guided by a calming voice coach) or uploads a pre-recorded sample.
  2. Analysis: The platform extracts audio features (envelope and energy) directly in the browser, ensuring rapid processing and privacy.
  3. Orchestration: The extracted waveform is sent to our backend where a team of specialized AI agents segment the breath cycle, compare it against historical baselines, and generate a clinical-style summary.
  4. Action: The user receives immediate feedback, including their Rhythm stability, Exhale Ratio, a personalized micro-intervention (e.g., "Lengthen your exhale by one count"), and a prompt for their next session.

Key Features

  • 3-Step 15s Capture: An intuitive, guided UI to capture high-quality audio samples effortlessly.
  • Waveform Analysis: Precise extraction of audio energy and envelope data to track the respiratory cycle.
  • Clinician-Style Summaries: Translates complex wave data into clear, non-diagnostic insights and actionable lifestyle coaching tips.
  • Historical Tracking: Visualizes and compares core breathing pillars—Rhythm, Exhale Ratio, Interruptions, and Holds—across past sessions to track progress or regression.
image

Powered By Multi-Agent Orchestration & Voice

RespiraSnap leverages cutting-edge AI orchestration and emotional voice generation to provide a seamless, intelligent user experience.

🧠 Backboard API

Backboard powers the core intelligence and stateful memory of RespiraSnap. We deeply integrated the backboard-sdk to utilize several of its key features:

1. Multi-Agent Orchestration Instead of relying on a single monolithic prompt, we use Backboard to divide the analysis into specialized, sequential tasks:

  • Segmentation Agent: Analyzes the raw audio feature stats to estimate precise inhale/hold/exhale timings.
  • Baseline & Trend Agent: Compares current snapshots against a user's historical data to track improvements or regressions over time.
  • Clinical Summary Agent: Consolidates the segmentation data into a readable, clinician-style overview.
  • Coaching & Follow-Up Agents: Generates personalized micro-interventions and creates a prompt for the user's next session. These agents run concurrently and stream their intermediate states (Queued → Running → Done) via Server-Sent Events (SSE) directly to the frontend, giving users a live, transparent view of the AI pipeline at work.
image image

2. Conversational Memory (RAG) A key feature of RespiraSnap is its ability to understand a user's baseline. We use Backboard's native Memory API to silently store past snapshot data (energy variance, rhythm stability, scores) as conversational memory attached to the user's device ID. When the Baseline & Trend Agent runs, it automatically retrieves this historical context to provide accurate, personalized trend analysis without us needing to build a complex RAG pipeline from scratch.

image

3. Thread & State Management Every analysis session belongs to an ongoing Backboard Thread. This ensures that the Follow-up and Coaching agents have immediate, conversational context of what the Segmentation and Baseline agents just concluded, maintaining continuity across the pipeline.

🗣️ ElevenLabs (Voice Coach)

To ensure high-quality audio capture, a steady breathing rhythm is essential. We integrated ElevenLabs to power the Voice Coach:

  • Guidance During Capture: The voice coach gently prompts the user ("Inhale...", "Hold...", "Exhale...") with a natural, calm, and human-like cadence.
  • Improved Data Quality: By actively helping users pace their breaths, the resulting audio signal is significantly cleaner, leading to far more accurate downstream segmentation by the multi-agent pipeline.

🏆 Hackathon Categories Targeted

RespiraSnap was engineered specifically to highlight the unique strengths of our sponsors:

1. Backboard.io - Best use of Backboard

We moved beyond a simple chatbot and used Backboard as a stateful AI operating system for a complex coordination problem (respiratory analysis).

  • Multi-Agent Orchestration: 4 distinct agents (Segmentation, Baseline, Clinical, Coaching) running sequentially on the same data.
  • Stateful Memory: We utilize Backboard's Memory API to persist audio feature profiles securely under device IDs. The Baseline agent retrieves this without us building a separate vector database.
  • Context Persistence: Health data shouldn't exist in a vacuum. By using threads and memory, RespiraSnap actually remembers if your rhythm was stable yesterday and adjusts its coaching today.

2. MLH x ElevenLabs - Best Project Built with ElevenLabs

We used ElevenLabs to build a fully autonomous, interactive voice companion.

  • Emotional Expressiveness: Instead of robotic prompts, the ElevenLabs voice acts as a calming presence, actively pacing the user's nervous system during the 15-second capture window. This dynamic audio experience directly improves the quality of the physiological data we collect.

3. [MLH] Best Hack Built with Google Antigravity

RespiraSnap was built entirely within the Agentic Development platform, Google Antigravity. We utilized its context-aware agent and natural language code commands to orchestrate our Next.js frontend, complex audio waveform processing, and the Backboard API integration seamlessly.


Tech Stack

  • Framework: Next.js 14, React 18, TypeScript
  • Visuals & Animations: CSS Modules, Framer Motion, React Three Fiber (3D visualizers)
  • AI/Agents: Backboard SDK (backboard-sdk)
  • Voice/TTS: ElevenLabs API
  • Database: MongoDB (Atlas/Local) for user accounts and snapshot history.

Getting Started

  1. Install dependencies:
npm install
  1. Set up environment variables: Create a .env.local file in the root directory:
BACKBOARD_API_KEY=your_backboard_live_or_sandbox_key
BACKBOARD_BASE_URL=https://app.backboard.io/api
ELEVENLABS_API_KEY=your_elevenlabs_api_key
MONGODB_URI=mongodb://localhost:27017/respirasnap
  1. Run the development server:
npm run dev
  1. Open the app: Navigate to http://localhost:3000 in your browser.

About

Project for Hack Canada 2026

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors