ContentEngine:Self Improving Agentic Content Creation System

ContentEngine is a self improving content generation agent system that learns and grows with the user through agentic memory, prompt self-optimization, and Reinforcement Learning

Comment

ContentEngine: Self-Improving Agentic Content Creation System

1. Core Motivation & Problem 💡

Category Description
Motivation It is fundamentally easier to generate authentic, high-quality content ideas in a conversation (like a podcast interview) than by facing the high cognitive load of a blank page. Top executives are interviewed for their best ideas—they don't write them cold.
Problem Current content creation methods force experts to be writers first, creating content that is slow, non-scalable, and often diluted. We solve the content ideation bottleneck.

2. What it Does / Core Features ✨

Feature Description
🎙️ AI Interview Agent Features a voice agent named Gammi that interviews the user in a podcast style to draw out unique insights and ideas.
📝 Dialogue-to-Article Pipeline Automatically takes the conversation transcript (or pasted dialogue) and runs it through a multi-agent loop to produce a polished article (currently Medium format).
🧠 Self-Improving Agent Learns and grows with the user through strategic memory, a prompt optimization loop, and Reinforcement Learning (RL) principles.

3. How We Built It / Architecture 🏗️

ContentEngine uses a multi-agent refinement loop leveraging cutting-edge tools:

Component Tool / Technology Role in the System
AI Layer / Agents Google Agent Development Kit (ADK) Powers the core multi-agent architecture (Initial Writer, Critic, Eval, Prompt Optimizer).
Voice Agent VAPI Powers the interactive voice agent, Gammi.
User Interface (UI) Copilotkit Used to quickly develop the entire application's user interface (AMAZING FRAMEWORK).
Observability Weave Captures insights, monitors agent performance, and provides visibility into the refinement loop.
Infrastructure Google Cloud Agent Engine Used for running and hosting the core agents.

4. Challenges & Accomplishments 🚧🏆

Category Details
Challenges Successfully integrating the Weave and ADK frameworks; fine-tuning Copilotkit to meet the necessary UI demands.
Accomplishments SHIPPING A WORKING PRODUCT in 10 hours; mastering and integrating ADK, Weave, and CopilotKit.

5. What's Next 🚀

Category Details
Future Vision To build the ultimate self-improving agent for content generation.
Immediate Next Steps Implement OpenPipe and ART for full Reinforcement Learning integration; implement Tavily for deep research capabilities to feed into the prompt optimizer agent.

Built With

  • agent-development-kit
  • cloud-run
  • copilotkit
  • google-adk
  • supabase
  • vercel
  • weave
Share this project:

Updates