← projects

PRJ-03REV 2026.10

Noterary

Ask plain-English questions about your own videos and recordings. Fully local, with every answer linked to the exact moment it was said.

tags
AI
stack
Python · faster-whisper · Qdrant · Ollama · LM Studio

Problem

I learn mostly from video, and I often need one specific two-minute explanation buried in hours of lectures. Scrubbing through recordings to find it gets old fast, and the same is true of meeting recordings.

What I built

  • Point Noterary at a folder. It notices new video or audio files and transcribes them automatically.
  • Ask a question like “What’s the intuition behind matrix multiplication?” and get an answer drawn from your recordings, with sources.
  • Click a source and the video plays from that exact moment, so you hear the answer in context instead of trusting a summary.
  • Filter by date to search only recent recordings, useful for questions like “what did we decide in Tuesday’s sync?”.
  • Everything runs locally: transcription (faster-whisper), vector search (Qdrant) and the LLM (Ollama or LM Studio). Recordings never leave your machine.

The demo uses 3Blue1Brown’s Essence of Linear Algebra series as the library.