Vale construir sua versão de VidNotes?

Build me a local video transcription and study-notes app to replace VidNotes. Requirements:

- A Python 3.12 app using FastAPI, Jinja templates, HTMX, and SQLite FTS5,
  served only on localhost:8787.
- The home screen accepts one YouTube URL or an MP4, MOV, M4A, or MP3 upload;
  use yt-dlp for the URL and ffmpeg to extract mono audio.
- Use the OpenAI Python SDK for timestamped speech-to-text and the Responses API;
  keep the API key in .env and split large audio into chunks with preserved offsets.
- Show queued, extracting, transcribing, generating, complete, and failed states. The
  result page has a player plus a searchable transcript with clickable timestamps.
- Generate a 5-bullet summary, key points, 10 flashcards, and a 5-question quiz
  from the completed transcript, with a button to regenerate each section.
- Store metadata and generated text in SQLite, media under ./data/projects/<uuid>/,
  and export the transcript and notes as Markdown, TXT, SRT, and JSON.
- No accounts and no telemetry. Everything stays local except yt-dlp downloads and
  OpenAI API calls, and each project keeps its original source URL.
- Out of scope: native apps, hosted web, cross-device sync, the Chrome extension,
  Instagram or TikTok extraction, and a public API, CLI, or MCP server.
- Include a README with setup, ffmpeg and yt-dlp installation, .env.example, data
  location, deletion steps, and a warning that source-site changes can break imports.