QUICKSTART.md

Quick Start Guide - Context Pruning with CrewAI

🚀 Get Started in 3 Steps

Step 1: Install Dependencies

cd /Users/saish/Downloads/Context_engineering/context_pruning
crewai install

Or manually:

uv pip install -e .

Step 2: Set Your API Key

Make sure your .env file contains:

GEMINI_API_KEY=your-actual-gemini-api-key-here
MODEL=gemini/gemini-flash-latest

Step 3: Run the Demo

crewai run

📊 What Will Happen

1. Retrieval Agent searches Lilian Weng's blog posts for information about reward hacking
2. Pruning Agent filters the content to remove irrelevant information
3. Response Synthesizer creates a comprehensive answer
4. Result saved to context_pruning_result.md

🎯 Example Query

Default query: "What are the types of reward hacking discussed in the blogs?"

Expected workflow:

[Retrieval Agent] Searching blog posts...
  → Retrieved ~15,000 tokens of content

[Pruning Agent] Filtering content...
  → Pruned to ~6,000 relevant tokens (60% reduction)

[Response Synthesizer] Creating answer...
  → Generated comprehensive markdown response

🔧 Customize Your Query

Edit src/context_pruning/main.py:

inputs = {
    'topic': 'hallucination in LLMs',
    'query': 'What causes hallucination in large language models?'
}

Then run again:

crewai run

📝 View Results

Check the generated file:

cat context_pruning_result.md

🐛 Troubleshooting

Import errors?

crewai install

API key issues?

• Verify GEMINI_API_KEY in .env
• Test with: echo $GEMINI_API_KEY

Want more verbose output?

• Agents already set to verbose=True in crew.py
• Check terminal for detailed execution logs

🎓 Learn More

• Compare with LangGraph implementation: ../how_to_fix_your_context/notebooks/04-context-pruning.ipynb
• Read full README: README.md
• Explore other context engineering techniques

💡 Pro Tips

1. First run is slower - Vector store initialization takes time
2. Check token counts - Watch the logs to see pruning effectiveness
3. Experiment with queries - Try different topics from the blog posts:
   • Thinking mechanisms in AI
   • Hallucination causes and solutions
   • Diffusion models for video
   • Reward hacking categories

Happy Context Pruning! 🎉