This tool was created for the Shameful Podcast competition. It performs the following tasks:
- Monitors the podcast RSS feed and downloads the latest episode automatically.
- Transcribes the episode using a local AI model running in PyTorch using CUDA cores to massively parallelize on the GPU.
- Identifies the secret word for each episode using fuzzy string searching.
- Determines the secret phrase by applying a coherence algorithm to find the most likely arrangement of the five words.
On the day of the competition, this tool solved the challenge just 9 seconds (!) after the episode was published.