Inspiration
Trying to keep up with the fast pace of news updates, spending too much time on news and podcasts.
What it does
Goes through thousands of news sources daily and telling you the most important condensed updates, so that you can stay up to date but also save your precious time.
How we built it
We have a scraping backend from RSS feeds and front pages. From these articles, we build a RAG vector database using qdrant to perform hybrid search with sparse and dense embeddings. For each user query, we retrieve relevant articles, generate a digest of basic facts agreed upon by all sources, and generate a debate using agents with specific beliefs and priors. Through multiple rounds of generation through gemini, we refine the arguments and summarize into a entertaining dialogue, which we use to generate an audio file with an open source huggingface model.
Challenges we ran into
Scraping was especially challenging because of paywalls and non standard tags across our website corpus. Audio generation was also difficult because of extremely low free token tiers and difficulty working with output files and dataset.
Accomplishments that we're proud of
Virtual debates, getting a MVP, retrieval quality.
What we learned
Coding assistants, very impressed by quality of arguments generated by gemini, and frontend stuff
What's next for Sincere Politics
Deliver to pilot users
Log in or sign up for Devpost to join the conversation.