Educational AI Pipeline Phase 2

Continuation of the Educational AI Pipeline project. Tasks include selecting a pilot PDF (algebra or science, high school level), comparing extraction methods across PyMuPDF, Marker, and Docling, cleaning and structuring content into a JSONL training dataset, running a first LoRA fine-tuning round on M4/MPS, evaluating agent performance on accuracy, context, information retrieval, and hallucination detection, and creating a repeatable pipeline from PDF to extraction, cleaning, JSONL, fine-tuning, adapter, and evaluation. Deliverables include extraction comparison, JSONL v1, LoRA adapter v1, evaluation report, and documented pipeline.