Engram is an organizer for what you've already learned. It turns handwritten notes — OneNote notebooks exported as opaque, unsearchable PDFs, or any PDF — into a searchable, interlinked Markdown vault for Obsidian. Built for people who take handwritten notes on an iPad, it transforms scattered ink on paper into a navigable second brain. Named after the neural engram, the physical trace a memory leaves behind, the project aims to give those traces a durable, structured home.
Engram — from ink to structure, from notes to memory.
A vision language model reads each page — handwriting, typed text, diagrams, and photos — and the pipeline parses every PDF, extracts pages, summarizes the key ideas and formulas while filtering noise, and writes the vault with original pages preserved as embedded PDFs. You choose the model: Claude Code (via login) for the strongest vision processing, or a local Qwen2.5-VL (7B / 32B via mlx-vlm) that runs fully on-device for privacy. Illegible sections are flagged rather than hallucinated, and incremental builds reprocess only changed or new notes.
The result is a dual structure: a folder hierarchy (Notebook → Section → Note) alongside a multi-layer concept graph that rolls ideas up through Concepts → Themes → Areas, revealing how topics connect across courses. Related notes are cross-linked automatically across many PDFs treated as one knowledge base, and a grounded chat lets you query or synthesize your vault with answers that are cited to your own notes and that clearly mark anything reaching beyond them.
Engram is an honest v0.2 prototype on Apple Silicon macOS that prioritizes privacy, reliability, and reproducibility over polish.