DEV Community

Farhan Mallik
Farhan Mallik

Posted on

NatureQuest Audio Explorer - Touch Grass with AI 🌲🎧

#devchallenge #hf26challenge #gemma #opensource

Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass Submission 🌿

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

🌟 What I Built

link web link
NatureQuest Audio Explorer is an open-source agentic web application engineered to solve digital screen fatigue. In a world of notifications and doomscrolling, NatureQuest gently nudges you to close your laptop, lace up your shoes, and literally touch grass.

Enter any city, neighborhood, or park, and NatureQuest orchestrates an autonomous AI agent that:

  1. Scouts real trails & parks near you using SerpApi.
  2. Crafts an inspiring 3-sentence sensory quest using the open-weight Gemma 2-9B model.
  3. Synthesizes a calming audio companion via ElevenLabs, complete with timeline scrubbers, volume controls, and offline download support.
  4. Preserves the quest into a global Community Vault powered by MongoDB Atlas.

🚀 Demo & Code

🏗️ How I Built It: The Agentic Workflow

Instead of a monolithic prompt, NatureQuest implements an autonomous agent architecture powered by Mastra:

graph TD
    A["User Input: Location (e.g. Kyoto Arashiyama)"] --> B["Mastra Agent Orchestrator (natureGuideAgent)"]
    B --> C["natureSearchTool (SerpApi)"]
    C -->|"Authentic outdoor spots & trailheads"| B
    B --> D["Gemma 2-9B Reasoning (Groq / Open-Weight)"]
    D -->|"Sensory-rich 3-sentence quest"| B
    B --> E["elevenLabsTtsTool (ElevenLabs)"]
    E -->|"Crystal-clear audio guide narration"| B
    B --> F["questMemoryTool (MongoDB Atlas)"]
    F -->|"Stores quest in Community Vault"| B
    B --> G["Interactive Next.js Player & Community Vault"]
[ User: "Kyoto Arashiyama" ]
            │
            ▼
┌────────────────────────────────────────────────────────┐
│            Mastra Agent Orchestration Loop             │
├────────────────────────────────────────────────────────┤
│ 1. natureSearchTool (SerpApi)                          │
│    Scouts authentic outdoor spots and trailheads       │
│                                                        │
│ 2. Gemma 2-9B Reasoning (Groq / Open-Weight)           │
│    Crafts sensory-rich 3-sentence quest                │
│                                                        │
│ 3. elevenLabsTtsTool (ElevenLabs)                      │
│    Generates crystal-clear audio guide                 │
│                                                        │
│ 4. questMemoryTool (MongoDB Atlas)                     │
│    Stores quest into global community vault            │
└────────────────────────────────────────────────────────┘
            │
            ▼
[ Interactive Next.js Player + Community Adventure Vault ]
Enter fullscreen mode Exit fullscreen mode

1. Agent & Tools Orchestration (Mastra)

We instantiated a genuine Mastra Agent (natureGuideAgent via @mastra/core/agent) coupled with modular tools via @mastra/core/tools (natureSearchTool, elevenLabsTtsTool, questMemoryTool). The orchestrator coordinates tool calls, validates inputs and outputs with Zod schemas, and handles graceful error boundaries.

2. Open-Weight Model Reasoning (Gemma via Groq)

We prompt Gemma 2-9B-IT using an intentional sensory framework:

  • Sentence 1 (Invitation): Direct call to step away from screens and visit a named trailhead.
  • Sentence 2 (Mindful Action): An actionable sensory cue (e.g. feeling cool pine bark, smelling damp soil, noticing wind patterns).
  • Sentence 3 (Perspective): A philosophical closing on how nature reconnects our mind.

3. Voice Companion & Clean Audio Streaming (ElevenLabs & Next.js)

Walking outside shouldn't mean staring down at your smartphone. NatureQuest converts the written quest into a soothing spoken narration using ElevenLabs Rachel voice. We designed a dual audio architecture: immediate base64 playback for newly synthesized adventures, plus a dedicated clean binary MP3 audio endpoint (GET /api/quest?id=...&audio=1) that delivers proper audio/mpeg streaming headers and HTTP caching without bloating JSON payloads. The frontend features a custom audio player with timeline seeking, waveform visualization, and an offline .mp3 download button so hikers can continue their quest even when cell reception drops.

4. Persistent Memory Layer (MongoDB Atlas)

Every quest is logged into MongoDB Atlas, creating a collective community vault of nature quests worldwide. Hikers can browse past quests, replay audio guides, and see what green spaces other explorers discovered.

5. Deployment & Reliability (Render & Sentry)

  • Render Web Service: Built with declarative infrastructure via render.yaml and a multi-stage Dockerfile.
  • API Guardrails: In-memory sliding-window rate limiter (15 requests/min per IP) to prevent DDoS and API quota exhaustion.
  • Sentry Integration: Real-time exception capture for front-end and back-end runtimes.

🌍 Why Open Innovation Matters

Open innovation and open-weight models like Gemma are foundational to the future of technology:

  1. True Offline Freedom in the Backcountry: Closed proprietary APIs require high-speed internet. When you hike into deep ravines or mountain trails where cell towers can't reach, closed models are useless. Because Gemma is open-weight, it can run locally on an edge device (e.g., using Ollama or llama.cpp on a phone or Raspberry Pi in a backpack), providing guidance anywhere on Earth.
  2. No Monopolistic Vendor Lock-in: Open models ensure developers aren't vulnerable to unexpected price hikes, sudden deprecations, or opaque content filtering on benign outdoor descriptions.
  3. Auditable & Fine-Tunable: Communities can adapt Gemma to regional ecology, indigenous botanical knowledge, and localized conservation efforts without asking permission from big-tech gatekeepers.

🏆 Prize Categories I am targeting

  • Best Use of Mastra: Modular createTool architecture, typed inputs/outputs, and autonomous tool orchestration.
  • Best Use of Gemma: Core generative intelligence powered by Google's Gemma 2-9B model.
  • Best Use of ElevenLabs: High-fidelity voice synthesis designed specifically for screen-free walking experiences.
  • Best Use of SerpApi: Live localized search discovering actual green spaces and parks.
  • Best Use of MongoDB Atlas: Scalable cloud document storage providing persistent adventure history.
  • Best Use of Render: Automated deployment pipeline via render.yaml infrastructure-as-code.

💚 Final Thought

Next time you feel overwhelmed by browser tabs or Slack notifications, remember: the outdoors is only a few steps away. Go outside, breathe deeply, and touch grass!

Top comments (1)

Collapse
 
farhan_mallik profile image
Farhan Mallik •

its new project for me , got to know about how "touch grass" - to get up from chair and see the outside world how important and difficult is will continue this further , thanks mlh for a theme i can relate to