ContentCraft – Multi-Agent AI Content Generation System

A production-grade agentic automation system that generates structured, machine-readable content pages from product data.

🚀 Features

• 6 Specialized Agents: Parser, Question Generator, FAQ, Product Page, Comparison, Output
• LangGraph Orchestration: StateGraph with parallel fan-out/fan-in execution
• 7 Logic Blocks: Including cross-block analyzer with impact scoring & risk assessment
• Quality Gates: validate_content_node with conditional routing (pass OR fail-fast)
• Hard Validation: Template validation strictly enforced - no fallback outputs
• Groq LLM Provider: Fast generation with llama-3.3-70b-versatile
• API Key Rotation: Multi-key support with automatic retry on rate limits
• Custom Template Engine: Class-based templates with validation
• Streamlit Demo UI: Interactive interface for content generation
• Machine-Readable Output: 3 JSON files (FAQ, Product, Comparison)
• 15+ FAQ Questions: With deduplication, scoring, and LLM regeneration loop
• Test Suite: 68 unit and integration tests included

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                    LangGraph StateGraph                     │
├─────────────────────────────────────────────────────────────┤
│  parse → generate_questions ─┬→ faq ──────────┐            │
│                              ├→ product_page ─┼→ validate  │
│                              └→ comparison ───┘    ↓       │
│                                              ┌─────┴─────┐  │
│                                              │  output   │  │
│                                              │    OR     │  │
│                                              │   END     │  │
│                                              └───────────┘  │
└─────────────────────────────────────────────────────────────┘

Workflow Features:

• Fan-out: 3 agents run in parallel after question generation
• Fan-in: All converge to validate_content_node
• Conditional Routing: Routes to output (success) OR END (fail-fast)

📁 Project Structure

├── agents/             # 6 agent implementations
├── logic_blocks/       # 7 blocks (benefits, safety, cross_block_analyzer, etc.)
├── templates/          # Template engine with hard validation
├── orchestrator/       # LangGraph workflow with quality gates
├── tests/              # 68 unit and integration tests
├── output/             # Generated JSON files
├── docs/               # Documentation
├── models.py           # Pydantic data models
├── config.py           # Groq config with API key rotation
├── utils.py            # Utility functions
├── app.py              # Streamlit UI
└── requirements.txt    # Dependencies

🛠️ Setup

1. Clone the Repository

git clone https://github.com/yk0007/kasparro-agentic-yaswanth-kuramdasu.git
cd kasparro-agentic-yaswanth-kuramdasu

2. Create Virtual Environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Configure Environment

cp .env.example .env
# Edit .env and add your API keys

Required API Key:

• GROQ_API_KEY - Groq API key for LLM generation

5. Run the Application

streamlit run app.py

🔧 LLM Configuration

Provider	Model	Use Case
Groq	llama-3.3-70b-versatile	All content generation

Note: This system uses Groq exclusively. Competitor products are fictional (no external search).

📖 Usage

1. Input Product Data: Enter product JSON or use text fields
2. Validate: Click "Validate JSON" to check the input
3. Generate: Click "Generate Content" to run the multi-agent workflow
4. View Results: See generated content in FAQ, Product, and Comparison tabs
5. Download: Click download buttons to save JSON files

✅ Running Tests

python3 -m pytest tests/ -v

All 68 tests verify:

• Assignment Requirements: Product parsing, question generation, templates, logic blocks
• Workflow Integration: LangGraph uses parallel fan-out/fan-in
• Workflow Resilience: Error handling, fail-fast behavior, recovery state
• No External Search: Only internal data used
• FAQ Generation: 15+ questions with deduplication and scoring

🔧 Example Input

{
  "name": "Your Product Name",
  "concentration": "Key differentiator",
  "skin_type": ["Target User 1", "Target User 2"],
  "key_ingredients": ["Feature 1", "Feature 2"],
  "benefits": ["Benefit 1", "Benefit 2"],
  "how_to_use": "Usage instructions",
  "side_effects": "Considerations or limitations",
  "price": "₹1999"
}

📊 Output Files

File	Description
faq.json	5+ categorized Q&A pairs
product_page.json	Complete product page content
comparison_page.json	Product A vs Competitor comparison

📚 Documentation

See docs/projectdocumentation.md for detailed system design and architecture.

🔑 Technologies

• Python 3.10+
• LangChain - Agent framework
• LangGraph - Workflow orchestration with StateGraph
• Groq API - Fast LLM (llama-3.3-70b-versatile)
• Pydantic - Data validation
• Streamlit - Web UI