Wunderkind: Multi-Agent Gen Z Magazine

Wunderkind is a fully automated, multi-agent AI magazine that creates daily news for teens. AI pitches the ideas, AI writes the stories, you make the final call.

Comment

🚀 Inspiration

Traditional news is often too dry or complex for teenagers, while social media feeds are full of misinformation. I wanted to build a solution that delivers engaging, age-appropriate, and factual news to Gen Z. However, I also knew that fully autonomous AI content can be risky. The inspiration for Wunderkind was to create a "Neuro Newsroom": a multi-agent AI system that does the heavy lifting of ideation and drafting, but keeps a human editor firmly in the loop for final approval.

🛠️ How I Built It

The project is a seamless orchestration of three main components:

  1. The Interface (Telegram): Each AI "author" is a dedicated Telegram bot, providing a familiar, conversational UI for the human editor.
  2. The Brain (OpenAI API): Powered by GPT-5.6 for text and DALL-E 3 for visuals. Each bot has a highly specialized system prompt (e.g., "Tech Expert", "Nature Enthusiast") tailored to speak to teenagers in an engaging, non-condescending tone.
  3. The Orchestrator (Make.com): Handles the workflow routing. When the editor requests ideas, Make.com triggers the specific agent to generate 5 topic pitches. Once a topic is selected, Make.com passes the context back to the same agent to write the full article and generate a matching image.

Note: The live demo operates in Russian to authentically serve our primary target audience, showcasing the model's strong multilingual capabilities. However, the architecture is entirely language-agnostic.

🧠 What I Learned

  • The power of specialized agents: A single generic prompt is good, but splitting the workload into distinct "persona" agents yields drastically higher quality and more consistent tone.
  • Human-in-the-loop is non-negotiable: For media and youth-oriented content, AI should be a co-pilot, not an autopilot. The editor's ability to request revisions is the most critical feature.
  • No-code/Low-code scalability: Make.com proved to be an incredibly robust tool for managing state and routing between Telegram and the OpenAI API without building a complex custom backend from scratch.

🚧 Challenges Faced

  • Context Management: Ensuring the AI remembered which of the 5 topics the editor selected when it was time to write the full article. Solved by carefully structuring the data payloads in Make.com.
  • Tone Calibration: Preventing the AI from sounding either too academic or too "cringe". This required multiple iterations of the system prompts, explicitly defining the "cool older sibling" persona.

Built With

Share this project:

Updates