AirWise-AI

AirWise AI uses live AQI, Elasticsearch Retrieval-Augmented Generation (RAG), and Gemini AI to deliver transparent, explainable, evidence-based public health decisions.

Comment

🚀 Inspiration

Air pollution affects millions of people every day, yet most people struggle to understand what the current AQI actually means for their daily activities. Existing weather or AQI apps usually display numbers without explaining how those conditions affect different people, while general AI assistants often provide answers without grounding them in trusted health guidance.

We wanted to build an AI system that combines live environmental data with trusted public health knowledge to help people make informed outdoor decisions. Instead of creating another chatbot, we focused on building an explainable AI decision support platform that is transparent about its reasoning and sources.

🌍 What it does

AirWise AI is an AI-powered public health decision support platform that combines live AQI, weather conditions, and trusted WHO/CPCB guidance to provide personalized health recommendations.

Users can:

  • 🌍 View real-time AQI, PM2.5, PM10, temperature, humidity, and wind speed.
  • 💬 Ask natural language questions such as:
    • Can I go for a walk this evening?
    • Should my child play outside today?
    • Do I need an N95 mask?
  • 👤 Select personalized health profiles including:
    • Child
    • Elderly
    • Asthma
    • Pregnancy
    • Outdoor Worker
  • 🧠 Receive AI-generated recommendations with:
    • Decision
    • Risk Level
    • Confidence Score
    • Recommended Actions
    • Decision Timeline
    • Evidence Citations
    • AI Provenance

Every recommendation is grounded in retrieved WHO and CPCB guidance instead of relying solely on the language model.

⚙️ How we built it

AirWise AI is built using a production-style Retrieval-Augmented Generation (RAG) architecture.

Tech Stack

  • Frontend: Next.js, React, TypeScript, Tailwind CSS
  • AI Model: Google Gemini
  • Search Engine: Elasticsearch
  • Live Data: Open-Meteo Air Quality API + CAMS
  • Knowledge Base: WHO & CPCB Health Guidelines

Architecture

User Question
        │
        ▼
Health Profile + Conversation Context
        │
        ▼
Live AQI & Weather Data
        │
        ▼
Intent Detection
        │
        ▼
Gemini Query Planner
        │
        ▼
Elasticsearch Multi-Stage Retrieval
        │
        ▼
Evidence Ranking & Compression
        │
        ▼
Gemini Grounded Reasoning
        │
        ▼
Explainable AI Response
        │
        ▼
Evidence + Provenance + Citations

Multi-Stage Retrieval Pipeline

  • Direct Lexical Retrieval
  • Query Expansion
  • Health Ontology Matching
  • Intent-Based Search
  • Evidence Ranking
  • Retrieval Compression

Only the highest-ranked evidence is passed to Gemini for grounded reasoning.

⚠️ Challenges we ran into

Building trustworthy AI required solving several engineering challenges beyond simply integrating an LLM.

Some of the biggest challenges included:

  • Improving retrieval quality for natural language questions.
  • Designing a multi-stage Elasticsearch retrieval pipeline.
  • Combining live AQI data with retrieved evidence.
  • Preventing hallucinations through grounded generation.
  • Handling API failures and model rate limits gracefully.
  • Making AI reasoning transparent through citations and provenance.
  • Personalizing recommendations using health profiles.
  • Maintaining performance while executing multiple retrieval stages.

🏆 Accomplishments that we're proud of

We are proud that AirWise AI evolved into a complete AI engineering project rather than a simple chatbot.

Some highlights include:

  • ✅ Live AQI Intelligence
  • ✅ Multi-City Support
  • ✅ Personalized Health Profiles
  • ✅ Multi-Stage Elasticsearch Retrieval
  • ✅ Retrieval-Augmented Generation (RAG)
  • ✅ Query Planning with Gemini
  • ✅ Explainable AI
  • ✅ AI Provenance
  • ✅ Evidence Ranking
  • ✅ Conversation Memory
  • ✅ Decision Timeline
  • ✅ Retrieval Evaluation Suite
  • ✅ Transparent Source Citations

📚 What we learned

This project taught us that building reliable AI systems requires much more than calling an LLM API.

We learned how Retrieval-Augmented Generation works in production, how Elasticsearch improves search quality, how live environmental data can be combined with AI reasoning, and why explainability and provenance are critical for trustworthy AI systems.

We also gained practical experience designing retrieval pipelines, integrating multiple APIs, evaluating search quality, optimizing latency, and engineering AI systems that prioritize transparency over hallucination.

🚀 What's next for AirWise AI

We plan to continue evolving AirWise AI into a comprehensive environmental health assistant.

Our roadmap includes:

  • 🔹 Hybrid Vector + Lexical Retrieval
  • 🔹 Official CPCB Station Integration
  • 🔹 Push Notifications for Dangerous AQI Levels
  • 🔹 Daily Personalized Health Reports
  • 🔹 Voice Assistant Support
  • 🔹 Mobile Application
  • 🔹 Multi-language Support
  • 🔹 Predictive AQI Forecasting
  • 🔹 Wearable Device Integration
  • 🔹 Expanded Environmental Health Knowledge Base

Our long-term vision is to build a trusted AI companion that helps people make healthier outdoor decisions through real-time environmental intelligence, evidence-based recommendations, and transparent AI reasoning.

Built With

Share this project:

Updates