Metrall | Devpost

Inspiration

The inspiration for the project came from the desire to make public information, such as city council meetings and news articles, more accessible and understandable to the general public. We wanted to bridge the gap between the overwhelming amount of public data and how citizens engage with it, making it easier for them to stay informed and contribute to civic discussions.

What it does

Metrall pulls in data from various sources such as city council meeting transcripts, news articles, and local events. It uses natural language processing and LLMs (Large Language Models) to summarize this data and present relevant information to users in a concise, easy-to-digest format. Users can query about local issues, political changes, or upcoming events, and the app responds with summaries and citations from its data sources.

How we built it

The app was built over a 24-hour hackathon using React/TypeScript on the front end and FastAPI with Python on the back end. We integrated LlamaIndex to handle data indexing and querying, and used Pinecone for vector storage. Data was pulled from various sources such as the News API and city council websites via public APIs, and we used Google Cloud to store and manage data. Front-end design was done in Figma

Challenges we ran into

We faced several challenges: Transcription quality: Ensuring the accuracy of transcripts from various sources was difficult, and handling the large video files for meetings was another hurdle Coding and integration issues: Integrating LlamaIndex with other components and ensuring the pipeline for data flow worked smoothly across all tools was challenging Team collaboration: Working remotely in a tight time frame meant coordinating between team members on tasks like embedding vectors, setting up databases, and handling design

Accomplishments that we're proud of

We successfully built an app that combines various data sources into a single searchable platform. We managed to set up a working end-to-end system that queries large amounts of public data and delivers it in an accessible way to users The integration of Pinecone, Google Cloud, and LlamaIndex into a cohesive pipeline was a technical highlight

What we learned

Teamwork: Effective communication and task management were crucial in ensuring each part of the project progressed smoothly Technical skills: We learned how to efficiently handle large data sets, manage ETL (Extract, Transform, Load) processes, and integrate complex API systems Product design: From user discovery to UX/UI design, we learned how to create an intuitive interface that encourages user engagement

What's next for Metrall

Expand data sources: We plan to expand the range of data sources beyond LA, including more cities and states, and integrate additional unstructured data. Refine AI summaries: We will work on improving the precision of the AI summarization to better suit different types of queries. Community features: Adding features like community discussions or a bulletin board where users can post events and news will further engage local communities.