Turn lecture PDFs into podcasts and flashcards—self-hosted

What it does PageLM ingests PDFs, Word docs, Markdown, and plain text, then uses LangChain and Langraph to generate interactive study aids: contextual chat, Cornell-style notes, flashcards, scored quizzes, and AI podcasts. It also bundles a homework planner, exam simulator, and voice transcription. The stack deploys via Docker and connects to whichever LLM you configure.

The interesting bit Rather than forcing a single vendor, the project supports Gemini, GPT, Claude, Grok, MiniMax, OpenRouter, and Ollama out of the box, streaming results over WebSockets. The default storage is plain JSON files instead of a mandatory vector database, so you can spin it up without first learning what pgvector is.

Key highlights

• Broad LLM support across Gemini, GPT, Claude, Grok, MiniMax, OpenRouter, and local Ollama
• Structured outputs: Cornell-style notes, non-overlapping flashcards, interactive quizzes, and podcast audio via Edge TTS, ElevenLabs, or Google TTS
• Real-time generation through WebSocket streaming for chat, notes, and podcasts
• File-based JSON persistence by default; optional vector database for retrieval
• Docker Compose configurations for both development and production

Caveats

• The CaviraOSS Community License allows personal and educational use but explicitly requires written permission for commercial deployment—read it before using at work.
• Mobile app support is listed as a future contribution area, so the interface is currently web-only.
• Default JSON storage may struggle with large document collections compared to a dedicated vector database.

Verdict Students, educators, and hobbyist self-hosters looking for a drop-in study tool should take a look. If you need an enterprise-grade learning management system with granular access control, this is not it.