Building an AI-powered personal knowledge assistant for the AMD Developer Hackathon: ACT III. The project ingests a user’s personal documents/notes, generates embeddings, and uses an open-weight LLM (e.g. Qwen2.5) served via vLLM on ROCm to answer questions and proactively surface non-obvious connections across the user’s knowledge base going beyond standard RAG chatbots. Will benchmark inference throughput on MI300X as part of the submission.