Skip to main content Skip to navigation
Back to Projects

Recipe Bot

A 2024 command-line MVP that turns Instagram cooking videos into Markdown recipes: download, transcribe with Whisper, check it is actually a recipe, then write it up with GPT.

PythonInstaloaderFFmpegWhisperOpenAI APIFirebaseprompt_toolkitPDM

The recipe is in the video. Nobody wrote it down.

Cooking reels say the quantities out loud and leave them out of the caption. Recipe Bot takes an Instagram post URL and gives back a written recipe in Markdown.

The pipeline

  1. Download: Instaloader fetches the post's video and caption, and FFmpeg pulls out the audio.
  2. Transcribe: OpenAI's open-source Whisper model (the small size, running locally) turns the audio into text.
  3. Check: before writing anything, gpt-4o-mini reads the transcript and caption and estimates how likely it is that the post contains a recipe at all. Below 85% it stops, so a dance video doesn't come back as a casserole.
  4. Write: the same model turns transcript plus caption into a structured recipe, saved as Markdown.

Audio and recipes are cached in Firebase Storage with metadata in Firestore, so a post is only downloaded and transcribed once, and there's a small terminal viewer built with prompt_toolkit for browsing saved recipes and opening them in an editor.

Status

A command-line MVP that Allison built over a weekend in November 2024 (53 commits), with a web app as the stated next step that never happened. It needs your own OpenAI key and Firebase project. The source is on GitHub, with no license.