Civic Compass

Meet the Candidate Page
Chat with Candidate Page
Landing Page
Current inference model
Proposed inference model

Inspiration

Civic Compass was born out of a commitment to enhancing informed civic participation. We aimed to leverage AI to cut through political noise and deliver the public clear, personalized civic insights.

What it does

Civic Compass serves as an AI-driven beacon in the complex landscape of political information. Our platform simplifies the political process, engages users through interactive learning, and fosters direct dialogue with AI representations of electoral candidates.

How we built it

To build Civic Compass, we followed a structured process:

Define the domain and datasets: We focused on U.S. election information, personalized experiences, and trust and accuracy as the key domains for our project.
Collect and preprocess data: We gathered relevant data from sources like the Civic Information API, election databases, and candidate websites.
Set up the infrastructure: We leveraged tools and platforms like modal.com, Nomic.ai, and the Anthropic API to build the necessary infrastructure for our project.
Implement the RAG architecture: We integrated the RAG model with a language model (such as Anthropic's Claude) and began fine-tuning the model using the RAFT technique on the selected datasets.
Develop the user interface: We designed and implemented a user-friendly web application to showcase the model's capabilities, including the Civic Information Feed and the interactive chatbot system.
Test and debug: We thoroughly tested and debugged the application, iterating on the design and functionality.

Challenges we ran into

Improving fine-tuning prompts: We had to craft precise and concise prompts to effectively fine-tune the language model. This required extensive experimentation and fine-tuning of the prompts to generate the desired level of detail and accuracy in the responses.
Integrating with the RAG chain: Integrating the RAG chain with the fine-tuned language model was a complex task, requiring careful coordination between the different components to ensure seamless information retrieval and generation.
Slow spool-up time for the API endpoint: The initial deployment of the API endpoint on modal.com resulted in a slow spool-up time, which could be overcome by increased usage or hosting the application on a dedicated server.

Accomplishments that we're proud of

We're proud to have created a Civic Information Feed and an interactive chatbot system that demystifies political information, making civic understanding accessible to all.

What we learned

We gained insights into the significance of domain-specific data, the complexity of AI integrations, and the paramount importance of user-centric design.

What's next for Civic Compass

Moving forward, we have several exciting plans for the development and expansion of Civic Compass:

Continuous model improvement: We aim to optimize the fine-tuning process further and explore more advanced language models to enhance the accuracy and depth of the information provided.
Expansion of data sources: We plan to integrate additional datasets and information sources to provide a more comprehensive and up-to-date Civic Information Feed.
Enhancing the chatbot capabilities: We will continue to refine the chatbot system, exploring ways to make the interactions more natural and engaging for users.
Implementing advanced gamification features: To foster deeper user engagement, we will introduce more sophisticated gamification elements, such as leaderboards, achievements, and interactive simulations.
Exploring multi-lingual support: To reach a wider audience, we aim to introduce multi-lingual capabilities, allowing users to access the Civic Information Feed and chatbot in their preferred languages.