FeedCast

Inspiration

Every day, we’re flooded with articles, videos, and news, but it’s hard to keep up without getting overwhelmed. We built FeedCast to turn that chaos into clarity: personalized, bite-sized audio summaries that let you stay informed while you’re on the go. Whether you’re commuting, walking to class, or grabbing coffee, FeedCast turns downtime into learning. What It Does FeedCast creates your own AI-generated news podcast — tailored to your interests. It curates trending stories, summarizes them into natural-sounding audio, and delivers them through an intuitive iOS app with real-time voice interaction. Users can interact with the podcast information, ask follow ups and clarifying questions. How We Built It We built a full-stack iOS app powered by a FastAPI backend. The iOS app uses SwiftUI for the interface and LiveKit for real-time voice chat. On the backend, we integrated Claude AI for script generation and fact-checking, OpenAI for conversation handling, and automated audio synthesis. Supabase manages authentication, database storage, and audio hosting. Our agent system orchestrates personalized content generation using prompt engineering for high-quality, engaging results. Challenges We Ran Into Getting LiveKit working smoothly for real-time audio was tougher than expected — a lot of debugging went into stable streaming. We initially tried FetchAI for the agent system but pivoted to a custom Claude-based pipeline for better control and content quality. Balancing personalization, accuracy, and natural flow in long-form AI speech took several prompt-engineering iterations. Accomplishments We’re Proud Of Built seamless real-time AI voice interaction with LiveKit Created an end-to-end system from user preferences → curated content → audio generation Designed synchronized text transcripts that follow along with playback Achieved natural, coherent AI-generated podcast segments through prompt refinement What We Learned We deepened our understanding of prompt engineering, real-time audio streaming, and multi-agent orchestration. We also learned how to integrate AI systems with a robust FastAPI backend and SwiftUI front end. What’s Next We plan to expand beyond news — adding educational content, research topics, and custom themes. The goal is to utilize this project as a way to make small quick actions based on your personal goals from adding a to do list to writing emails.