ENVI

Environment, simplified

Comment

🌱 ENVI — Environmental Intelligence Powered by Gemini

🌏 Inspiration

The inspiration for ENVI stems from the escalating environmental crisis in India. In a country where hazardous smog and record-breaking heatwaves are becoming the “new normal,” we realized that people weren’t suffering from a lack of data, but a lack of intelligence.

Raw AQI numbers don’t tell a mother if her child can play outside.

We saw an opportunity to use Google Gemini’s multimodal reasoning to transform cold telemetry into empathetic, life-saving advice.

🚀 What It Does

ENVI is a Personal Environment Architect that uses Gemini to bridge the gap between sensing and acting:

✅ Gemini AI Advisor

  • A context-aware advisor powered by Gemini 3 that analyzes:

    • Health profile (asthma, activity level)
    • Real-time environmental data

  • Provides custom activity windows.

✅ Recursive ENV Score

A unified safety metric weighted by AI analysis:

$$ S_{env} = \sum_{i=1}^{n} (w_i \cdot \text{Metric}_i) + \text{AI_Risk_Adjustment} $$

This converts multiple environmental variables into a single, actionable score.

✅ Green Cover Reports

  • Uses Gemini with Google Search Grounding
  • Powered by Google Antigravity for advanced spatial and environmental reasoning
  • Generates dynamic City Intelligence Reports

  • Identifies:

    • Reforestation zones
    • Local plant species
    • Canopy gaps

✅ Predictive Patterns

  • Analyzes historical trends using Gemini 3
  • Predicts the cleanest hour of the day
  • Helps users plan activities around air quality

🛠️ How We Built It

At the core of ENVI is a Gemini Orchestration Layer, developed and tested using Google AI Studio.

🔹 Model Fallback Chain

We implemented smart routing across Gemini models:

  • Gemini Flash → Fast, cost-efficient processing
  • Gemini Pro → Complex reasoning and deep-dive reports
  • Gemini 3 → Multimodal and predictive intelligence

🔹 Search Grounding & Synthesis

  1. Fetches live data using googleSearch
  2. Enhances spatial understanding via Google Antigravity
  3. Passes results to a secondary Gemini instance
  4. Converts unstructured text into structured JSON

🔹 Structured Output

We leveraged:

{
  "responseMimeType": "application/json"
}

This ensures perfectly formatted data for maps and charts.

🔹 Cloud Infrastructure

  • Fastify backend
  • AWS DynamoDB caching
  • TTL-based insight storage
  • High performance + freshness

🚧 Challenges We Ran Into

⚠️ Chaining the Brains

  • Orchestrating Gemini 3, Pro, and Flash models
  • Preventing context drift
  • Heavy prompt engineering in Google AI Studio

⚠️ Embedding Green Cover Intelligence

  • Required advanced grounding with Antigravity
  • Avoided listing only famous parks
  • Focused on underserved zones

⚠️ Latency vs Intelligence

Balancing:

  • Deep reasoning (Gemini Pro & 3)
  • Instant responses (Flash)

⚠️ Work Stress

Building:

  • Custom design system
  • AI pipelines
  • AWS integrations

…in parallel was intense.

🏆 Accomplishments We’re Proud Of

We built a System 2 AI application, not just a wrapper.

Gemini is deeply embedded in the logic flow.

The moment Gemini—powered by Antigravity and Gemini 3—identified a low-income street in Bangalore as a High-Priority Reforestation Zone using live canopy data, we knew we had something special.

📚 What We Learned

📌 Grounding Is Everything

AI is only as good as the data it sees.

Search grounding and Antigravity turned a chatbot into a researcher.

📌 AI as a Logic Layer

LLMs can transform (N) variables into a single, human-readable safety score.

📌 Shipping Consistency

A polished product matters. AI should feel like a feature, not a gimmick.

📌 Rapid Prototyping with AI Studio

Google AI Studio enabled fast experimentation, evaluation, and deployment of multi-model workflows.

🔮 What’s Next for ENVI

Gemini will power the next phase:

🌍 Global Trend Synthesis

  • Compare 100+ cities
  • Identify “Twin Cities”
  • Share scalable solutions

💬 AI-Moderated Community Hubs

  • Summarize citizen feedback
  • Convert discussions into reports
  • Assist local governments

🌱 Green Event Integration

  • Match reforestation zones
  • Link with volunteer events
  • Automate planning using Gemini’s reasoning capabilities

Built With

Share this project:

Updates