EcoCT

Inspiration: We Just Want to Breathe Easy 😮‍💨

Let's be honest: our team is over the pollution struggle. Seriously, some of us grew up breathing this air, and one of us has asthma—so we're personally invested in making things better.

That's why we built EcoCT.

We decided that managing your environmental safety shouldn't be a chore. Why check five different apps when you could check one? EcoCT is your comprehensive helper, handling everything from air quality to marine tides to simple rain or shine weather.

Powered by AI-agents on Agentverse (Fetch.ai), we've put all the essential info in one convenient spot. Think of us as your peace-of-mind portal. We're just trying to make it easier for everyone—including ourselves—to stay safe and enjoy the day!

What it does

EcoCT is a conversational environmental dashboard:

Natural language Q&A about environmental concerns (air quality, marine conditions, climate change, etc.)
Live data (Open‑Meteo, OpenAQ) with seamless CSV fallback for offline demos
Instant insights: concise answers, KPI tiles, and charts
Safety and planning: sea state and small‑craft advisory cues; AQI categories and trends
Prediction: predict the near future

How we built it

Frontend

Streamlit App (Python) – interactive chat interface + real-time environmental dashboard
Responsive UI with insight cards, AQI history chart, and geo map
Reusable components for query input, result summaries, and agent messages
Plotly for time-series visualizations (AQI, PM2.5, wind correlation)
Mapbox for location-based pollutant overlays and hotspot mapping
Minimal UX latency via async API calls to the agent network

Backend

Python backend using uAgents runtime for autonomous messaging
Inference + Orchestration with Fetch.ai Agentverse for agent discovery & routing
REST bridge using FastAPI to expose /ask and /insights endpoints
Typed messaging via Pydantic schemas for safe agent communication
Robust fallback caching to handle API rate limits or offline mode

🤖 Fetch.ai Agent Network (Core Logic)

We designed a multi-agent architecture using Fetch.ai where each agent has a focused role and communicates autonomously:

Agent	Purpose
AirAgent	Open‑Meteo Air Quality API, multi‑pollutant, AQI categories
MarineAgent	Open‑Meteo Marine API, waves and sea state
Climate comparison	Open‑Meteo Climate API, baseline vs recent
Frontdoor agent	CSV ingestion, anomaly hints, and micro‑actions
Forecast Agent	Predicts short-term air quality trends using historical data and Gemini-powered reasoning to generate localized AQI forecasts

AirAgent (Open‑Meteo Air Quality API, multi‑pollutant, AQI categories)
MarineAgent (Open‑Meteo Marine API, waves and sea state)
Climate comparison (Open‑Meteo Climate API, baseline vs recent)
Frontdoor agent (CSV ingestion, anomaly hints, and micro‑actions)

Challenges we ran into

We tried making two agents converse, one agent calling an LLM agent, but it failed due to incorrectly formatted input data. To address the problem, we use Gemini instead of the LLM agen from fetch.ai.
This is the first time we use fetch.ai, we decided to create a more complicated agents. We spent a lot of time perfecting the APIs and parsing data.

Accomplishments that we're proud of

Clean end‑to‑end experience: upon asking a question, users can see the map and trend in one flow
Robust fallbacks: runs offline with sample CSV, gracefully degrades if APIs fail
Climate comparison integration with meaningful, confidence‑scored summaries
Thoughtful visual polish: themed UI, helpful placeholders, and debug expanders

What we learned

Keep agent modules composable: fetch + shape data; defer UI logic to the app
Normalize chart inputs early; add categorical fallbacks for non‑temporal comparisons