Inspiration

The rise of autonomous AI agents has revolutionized how tasks are performed—but most AI applications still rely on isolated functionalities. We wanted to explore how a coordinated team of specialized AI agents could collaborate to create high-quality, multi-format content (text, code, image prompts) from just a single topic. Our goal: Build a smart, end-to-end content generation pipeline that automates research, writing, refinement, coding, and visualization with minimal human input.

What it does

ContentCrafter AI is a multi-agent system that takes a single topic as input and produces:

A structured content plan

In-depth research insights

Long-form, informative content

Refined, well-formatted output (tone/language customization)

Python code examples related to the topic

Descriptive image prompts (for tools like DALL·E or Midjourney)

All of this is powered by six collaborating agents:

Planner Agent – Outlines the strategy

Research Agent – Gathers factual data

Content Generator – Writes long-form content

Refinement Agent – Polishes tone, clarity, grammar

Code Agent – Adds sample Python code

Image Agent – Crafts AI art prompts

How we built it

Framework: Python with ADK structure

UI: Streamlit for easy interaction

LLM Backend: GROQ API with Mixtral (and fallback to LLaMA 3 70B when Mixtral was deprecated)

Modules: Modular Python files per agent (planner, research, etc.)

Orchestration: A central OrchestrationAgent manages the flow from topic input to output delivery

Dotenv: For managing API keys securely

Challenges we ran into

🧠 GROQ Model Deprecation: Our initial model (Mixtral) was deprecated mid-development, so we reworked the system to support fallback models.

🔄 Agent Coordination: Designing agents to pass meaningful context between each other without redundancy or hallucination required multiple prompt iterations.

💡 Pydantic Modeling Errors: ADK’s BaseModel conflicts caused initialization issues, which were resolved by restructuring constructors.

🧪 Testing Streamlit Locally: Some environments failed due to API limits or misconfigurations, so fallback logging and verbose output were added for debugging.

Accomplishments that we're proud of

Created a functioning multi-agent pipeline with fully modular agents

Successfully integrated GROQ API into a custom LLM orchestration flow

Designed a clean, intuitive UI that makes advanced multi-agent tech usable for everyday users

Adapted to last-minute model deprecation and kept the project working

What we learned

How to build agentic systems using Pydantic, Streamlit, and modular design

Fine-tuning prompts for agents to reduce hallucinations and increase accuracy

Deep understanding of GROQ API limitations, model behavior, and fallback planning

How to manage LLM-driven workflows in an extensible and maintainable way

What's next for ContentCrafter AI

Add a Blogger publishing agent to auto-publish refined content

Integrate voice-over or narration generation for accessibility

Enable visualization support, turning data into charts/images automatically

Create multi-language pipelines using translation agents

Package as a SaaS microservice for content marketers, educators, and bloggers

Built With

Share this project:

Updates