hnfm: building a local-first AI podcast generator for Hacker News articles (briancaffey.github.io)

The hnfm project is an innovative local-first AI podcast generator that transforms Hacker News articles into multimedia podcast-style videos—complete with narration, imagery, and subtitles—entirely on consumer hardware without relying on cloud services. Developed as an entry for the OpenAI Open Model Hackathon, hnfm showcases the power of open-source AI models running on a single RTX 4090 GPU, integrating a custom large language model (gpt-oss-20b) for content summarization and scriptwriting, a realistic text-to-speech system (nari-labs/dia), a diffusion-based image generator (InvokeAI with Flux Krea), and WhisperX for high-precision speech recognition and subtitle alignment. What sets hnfm apart is its end-to-end modular pipeline that fetches Hacker News posts using public APIs, cleans and indexes article content, generates personalized two-speaker podcast scripts tuned for reasoning complexity, and outputs synchronized audiovisual episodes. This approach enables richer, more accessible content consumption tailored to personal interests, blending text, audio, and visuals without exposing data to external platforms. Technically, the system leverages FastAPI, Celery, Redis, LangChain, and Dockerized services for orchestration, allowing scalable asynchronous processing and fine-grained resource management. For the AI/ML community, hnfm exemplifies the potential of local, open-source AI workflows that emphasize ownership, creative control, and prosumer content generation. It challenges the prevailing cloud-dependent paradigm, proving that complex multimedia AI applications can be feasible and performant on personal devices. By open-sourcing the project, hnfm invites collaboration and further exploration of fully self-hosted AI media creation, representing a significant step toward democratizing advanced AI tools outside of proprietary ecosystems.