Skip to main content

​
How it works

KommtKevinOnline solves a real problem for Papaplatte’s community: Kevin mentions his next stream plans at the end of each broadcast, but this information isn’t easily accessible. The app automatically extracts these announcements using AI-powered VOD analysis.

​
The workflow

1

Twitch VOD retrieval

The system monitors Papaplatte’s Twitch channel using the Twitch API. When a stream ends, it retrieves the VOD (Video on Demand) and extracts the last few minutes of audio—the segment where Kevin typically announces his next stream.
2

Speech-to-text transcription

The audio from the VOD is processed through OpenAI’s Whisper model, which converts Kevin’s German speech into accurate text transcriptions. Whisper handles background noise, music, and various speaking styles effectively.
3

ChatGPT classification

The transcribed text is analyzed by ChatGPT to identify stream announcements. The AI model is specifically prompted to:
  • Detect if Kevin mentioned when he’ll stream next
  • Extract specific dates, times, or relative timeframes (“tomorrow”, “in two days”)
  • Classify the intent: Is he streaming soon? Taking a break? Uncertain?
4

Database storage and display

The prediction results are stored in a PostgreSQL database and immediately displayed on the website. Users see a simple Yes/No answer with the predicted stream time, updated automatically after each VOD analysis.

​
Technology stack

KommtKevinOnline is built on a modern web stack optimized for real-time data and AI processing:
  • Nuxt - Server-side rendering framework built on Vue.js, powering both the frontend UI and backend API
  • PostgreSQL - Relational database storing VOD analysis results, predictions, and alert subscriptions
  • Whisper - OpenAI’s speech recognition model that transcribes German audio with high accuracy
  • ChatGPT - Large language model that classifies transcribed text to extract stream announcements
  • Twitch API - Provides real-time stream status, VOD access, and channel information
  • Vuetify - Material Design component library for the user interface
  • 7TV - Emote integration for authentic Twitch community experience

​
Why this approach works

Manual checking of VODs is time-consuming and impractical for thousands of viewers. By automating the process with AI:
  1. Accuracy: Whisper provides near-human transcription quality for German speech
  2. Speed: Analysis completes within minutes of a stream ending
  3. Consistency: Every VOD is analyzed using the same methodology
  4. Accessibility: Information that was buried in hours of content is now instantly available
The AI models are specifically tuned to handle Kevin’s speaking style, colloquialisms, and the typical phrases he uses when announcing streams.

​
Flow diagram

The complete data flow from stream end to prediction display: Application Flow This diagram shows how data moves through the system: from Twitch VOD retrieval, through Whisper transcription and ChatGPT analysis, into the PostgreSQL database, and finally to the web interface where users see the prediction.

Build docs developers (and LLMs) love

Get started for freeTalk to us