LangGraph Research Assistant 🔍

A simple iterative research agent built with LangGraph that searches the web and refines answers automatically.

Built for Google Colab | Uses Groq (Free) | Real Web Search | No API Cost for Search

---

🎯 What Does This Do?

This notebook demonstrates LangGraph - a framework for building AI agents that can:

• ✅ Loop and retry (not just linear chains)
• ✅ Make decisions (should I search again?)
• ✅ Refine queries automatically
• ✅ Prevent infinite loops (max 3 iterations)

Example Flow:

Question → Search → Bad Answer? → Refine Query → Search Again → Good Answer ✓

---

🚀 Quick Start (Google Colab)

1️⃣ Open in Colab

Click here →

2️⃣ Get Your Free Groq API Key

1. Go to groq.com
2. Sign up (it's free!)
3. Create an API key
4. Copy it

3️⃣ Add API Key to Colab

# In Colab: Click 🔑 (Secrets) on left sidebar
# Add: Name = "GROQ", Value = "your-api-key-here"

4️⃣ Run All Cells

Press Runtime > Run all and watch it work!

---

📦 What Gets Installed

# This happens automatically in Cell 1:
!pip install langgraph langchain_groq langchain_community -U ddgs

No OpenAI required! Uses:

• Groq → Free fast LLM (llama-3.3-70b)
• DuckDuckGo → Free web search (no API key!)

---

🔧 How It Works

1. State (Shared Memory)

question: str           # Original question
search_query: str       # Refined search query
search_results: str     # What we found
answer: str            # Final answer
needs_more_info: bool  # Should we loop?
iteration_count: int   # Safety counter (max 3)

2. Nodes (Workers)

• analyze_question → Creates/refines search query
• search_web → Searches DuckDuckGo
• generate_answer → Uses LLM to answer

3. Edges (Flow Control)

START → analyze → search → generate → ⚡ Decision Point
                              ↑              ↓
                              └─────(loop)───┘  or END

The Magic: Conditional edge decides:

• If needs_more_info == True AND iterations < 3 → Loop back
• Otherwise → Stop and return best answer

---

💡 Example Output

❓ Question: How many episodes in One Piece anime?

🔍 Searching for: How many episodes in One Piece anime?
✅ Found results (length: 1237 chars)
🤖 Generating answer (iteration 1)...
⚠️  Answer incomplete, needs more research
🔄 Looping back for more research (iteration 1/3)

🔍 Searching for: One Piece anime total episodes count 2024
✅ Found results (length: 760 chars)
🤖 Generating answer (iteration 2)...
✅ Answer complete!

FINAL ANSWER:
The One Piece anime has released more than 1150 episodes as of November 2024.

Total iterations: 2

---

🎨 Customize It

Change the Question

question = "What are the latest AI trends?"  # Your question here

Adjust Max Iterations

# In should_continue_research function:
if state["needs_more_info"] and state["iteration_count"] < 5:  # Change 3 to 5

Use Different LLM

llm = ChatGroq(
    model="mixtral-8x7b-32768",  # Faster, cheaper
    # or "llama-3.3-70b-versatile"  # Smarter, slower
)

---

🧠 Why LangGraph vs LangChain?

Feature	LangChain	LangGraph
Flow	Linear (A→B→C)	Graph (loops, branches)
Decisions	No	Yes
Retries	Manual	Built-in
Use Case	Simple RAG	Complex agents

Example:

• LangChain: Question → Retrieve → Answer ✅ Good for simple Q&A
• LangGraph: Question → Search → Bad? → Retry → Good! ✅ Better for complex tasks

---

📊 Real-World Use Cases

1. Research Assistant - Searches until it finds complete answer
2. Customer Support - Routes questions to right department
3. Code Reviewer - Analyzes → Finds issues → Suggests fixes → Loops
4. Content Writer - Research → Outline → Write → Self-edit → Publish

---

🛠️ Troubleshooting

Error: "NameError: name 'GROQ' is not defined"

• Solution: Add your Groq API key in Colab Secrets (🔑 icon on left)

Search returns empty results

• DuckDuckGo might be rate-limiting
• Try again in a few seconds

Answer says "NEED_MORE_INFO" 3 times

• Question might be too specific or recent
• Try rephrasing the question

---

📚 Learn More

• LangGraph Docs
• Groq API Docs
• LangChain Docs

---

🤝 Contributing

Found a bug? Have an improvement?

1. Fork this notebook
2. Make changes
3. Share your version!

---

📝 License

MIT License - Use freely in your projects!

---

⭐ If this helped you understand LangGraph, give it a star!

Built with ❤️ for beginners learning AI agents

---

🎓 Next Steps

After this, try:

1. Add memory (save conversation history)
2. Use multiple agents (researcher + writer)
3. Add human-in-the-loop (ask user for clarification)
4. Connect to your own data (RAG with iteration)