CarbonScope

Where AI maps Emissions into Meaning

Comment

Inspiration

The ever-growing urgency of the climate crisis and the complexity of carbon emission data scattered across countries, industries, and real-time news sparked the idea. I wanted to create a system that not only analyzes structured datasets but also weaves in timely insights from global news. This was also a natural intersection of my interests in AI, data visualization, and meaningful storytelling.

What it does

1. Analyzes Emissions by Country and Sector Two specialized agents analysis_country_agent and analysis_sector_agent—dig into carbon emissions data from various countries and industrial sectors, helping users uncover meaningful trends and differences.

2. Mines and Interprets Relevant News A third agent, news_analyst, retrieves recent news articles related to emissions or environmental changes and derives insights, summaries, or data cues using embedding-based similarity search.

3. Orchestrates Multi-Agent Collaboration A central manager agent oversees the entire process. It delegates tasks intelligently to the right sub-agent based on user queries

4. Supports Interactive Insights and Visualization both the sub agents analysis_country_agent and analysis_sector_agent are capbale of providing data visualization within the chat

How we built it

a. Architecture: The system is built using a modular multi-agent setup using Python, where a manager agent delegates tasks to three sub-agents: analysis_sector_agent – analyzes data across industry sectors. analysis_country_agent – compares countries’ emissions footprints. news_analyst– extracts and summarizes emissions-related headlines.

b. Data ingestion: I have created a MongoDB database using MongoDB Atlas to store the climate TRACE data along with the vector embeddings for creating vector search indexes on the names of the countries and sector.

c. Orchestration: The agents interact through clearly defined APIs, and the manager uses a delegation strategy to invoke the appropriate tools based on the request.

d. Deployment: I containerized the application using Docker and deployed it via Google Cloud Run, making it accessible for real-time interaction.

Challenges we ran into

  1. Vector Index Tuning: Generating meaningful and lightweight embeddings that balance speed and precision took several iterations.

  2. Deployment Debugging: Deploying the agent using Google Cloud run , was pretty much difficult and challenging , because all the errors during deployment were unknown and new.

  3. Incorporating visualizations : Since it was completely new for me to develop an agent from scratch , so getting the right resource on how to show charts within the chat was not at all easy.

Accomplishments that we're proud of

Designed a Fully Modular Multi-Agent System

Deployed the Project Using Google Cloud Run

Bridged Data with Visualization Tools

Pushed Through Complex Debugging Challenges

What we learned

a. Agent Collaboration: I deepened my understanding of multi-agent design

b. MongoDB + Vector Search: I honed skills connecting MongoDB datasets with embeddings, enabling faster similarity searches using vector indexes

c. Real-time Pipeline Design: I explored streaming-like models to keep emissions analysis dynamic and responsive to unfolding news.

What's next for CarbonScope

1. Live Data Streams & Real-Time Updates Integrate with APIs like OpenAQ or satellite feeds to enable near real-time tracking of CO₂ emissions, allowing your agents to respond to ongoing events more dynamically.

2. Topic-Aware News Digest Extend news_analyst to generate daily or weekly digests with embeddings and clustering techniques, grouped by sector or geography, complete with auto-generated titles.

3. Collaboration API Open CarbonScope as a platform—let other researchers or civic groups query or plug in their tools through an API. 4. Geospatial Insight Layer Augment the system with maps using libraries like Folium or kepler.gl to visually explore emission hotspots.

Built With

Share this project:

Updates