A personal AI knowledge assistant puts your entire information environment at your fingertips. Architecture: document collection (Obsidian notes, PDFs, web clips, emails) -> automated ingestion pipeline -> vector store -> LLM Q&A interface. Implementation with LlamaIndex: 1) Document sources: Obsidian Reader for markdown notes, PyMuPDF for PDFs, Jina Reader API for web pages. 2) Processing: recursive character splitting (chunk_size=512, chunk_overlap=50), metadata preservation (source, date, title). 3) Index: Qdrant local instance (docker run -p 6333:6333 qdrant/qdrant). 4) Querying: RetrieverQueryEngine with HyDE (Hypothetical Document Embeddings) for better recall. Personal use features: 1) Connection discovery: "What notes relate to this concept?" using semantic search across all documents. 2) Daily briefing: morning query synthesizing recent notes, flagging topics for review. 3) Citation tracking: when answering, show source documents and specific passages. 4) Spaced repetition integration: identify notes not accessed in 30+ days for review. Privacy: run entirely locally with Ollama (Llama 3.1 8B) for zero data transmission. 8B models handle most knowledge retrieval tasks acceptably. Building workflow: ingest new documents automatically via watched folder, incremental updates to existing index.