A Step-by-Step Implementation Tutorial for Building Modular AI Workflows Using Anthropic’s Claude Sonnet 3.7 through API and LangGraph

from getpass import getpass
import os


anthropic_key = getpass("Enter your Anthropic API key: ")


os.environ["ANTHROPIC_API_KEY"] = anthropic_key


print("Key set:", "ANTHROPIC_API_KEY" in os.environ)

import os
import json
import requests
from typing import Dict, List, Any, Callable, Optional, Union
from dataclasses import dataclass, field
import networkx as nx
import matplotlib.pyplot as plt
from IPython.display import display, HTML, clear_output

try:
    import anthropic
except ImportError:
    print("Installing anthropic package...")
    !pip install -q anthropic
    import anthropic


from anthropic import Anthropic

@dataclass
class NodeConfig:
    name: str
    function: Callable
    inputs: List[str] = field(default_factory=list)
    outputs: List[str] = field(default_factory=list)
    config: Dict[str, Any] = field(default_factory=dict)

class LangGraph:
    def __init__(self, api_key: Optional[str] = None):
        self.api_key = api_key or os.environ.get("ANTHROPIC_API_KEY")
        if not self.api_key:
            from google.colab import userdata
            try:
                self.api_key = userdata.get('ANTHROPIC_API_KEY')
                if not self.api_key:
                    raise ValueError("No API key found")
            except:
                print("No Anthropic API key found in environment variables or Colab secrets.")
                self.api_key = input("Please enter your Anthropic API key: ")
                if not self.api_key:
                    raise ValueError("Please provide an Anthropic API key")
       
        self.client = Anthropic(api_key=self.api_key)
        self.graph = nx.DiGraph()
        self.nodes = {}
        self.state = {}
   
    def add_node(self, node_config: NodeConfig):
        self.nodes[node_config.name] = node_config
        self.graph.add_node(node_config.name)
        for input_node in node_config.inputs:
            if input_node in self.nodes:
                self.graph.add_edge(input_node, node_config.name)
        return self
   
    def claude_node(self, name: str, prompt_template: str, model: str = "claude-3-7-sonnet-20250219",
                   inputs: List[str] = None, outputs: List[str] = None, system_prompt: str = None):
        """Convenience method to create a Claude API node"""
        inputs = inputs or []
        outputs = outputs or [name + "_response"]
       
        def claude_fn(state, **kwargs):
            prompt = prompt_template
            for k, v in state.items():
                if isinstance(v, str):
                    prompt = prompt.replace(f"{{{k}}}", v)
           
            message_params = {
                "model": model,
                "max_tokens": 1000,
                "messages": [{"role": "user", "content": prompt}]
            }
           
            if system_prompt:
                message_params["system"] = system_prompt
               
            response = self.client.messages.create(**message_params)
            return response.content[0].text
       
        node_config = NodeConfig(
            name=name,
            function=claude_fn,
            inputs=inputs,
            outputs=outputs,
            config={"model": model, "prompt_template": prompt_template}
        )
        return self.add_node(node_config)
   
    def transform_node(self, name: str, transform_fn: Callable,
                      inputs: List[str] = None, outputs: List[str] = None):
        """Add a data transformation node"""
        inputs = inputs or []
        outputs = outputs or [name + "_output"]
       
        node_config = NodeConfig(
            name=name,
            function=transform_fn,
            inputs=inputs,
            outputs=outputs
        )
        return self.add_node(node_config)
   
    def visualize(self):
        """Visualize the graph"""
        plt.figure(figsize=(10, 6))
        pos = nx.spring_layout(self.graph)
        nx.draw(self.graph, pos, with_labels=True, node_color="lightblue",
                node_size=1500, arrowsize=20, font_size=10)
        plt.title("LangGraph Flow")
        plt.tight_layout()
        plt.show()
       
        print("\nGraph Structure:")
        for node in self.graph.nodes():
            successors = list(self.graph.successors(node))
            if successors:
                print(f"  {node} → {', '.join(successors)}")
            else:
                print(f"  {node} (endpoint)")
        print()
   
    def _get_execution_order(self):
        """Determine execution order based on dependencies"""
        try:
            return list(nx.topological_sort(self.graph))
        except nx.NetworkXUnfeasible:
            raise ValueError("Graph contains a cycle")
   
    def execute(self, initial_state: Dict[str, Any] = None):
        """Execute the graph in topological order"""
        self.state = initial_state or {}
        execution_order = self._get_execution_order()
       
        print("Executing LangGraph flow:")
       
        for node_name in execution_order:
            print(f"- Running node: {node_name}")
            node = self.nodes[node_name]
            inputs = {k: self.state.get(k) for k in node.inputs if k in self.state}
           
            result = node.function(self.state, **inputs)
           
            if len(node.outputs) == 1:
                self.state[node.outputs[0]] = result
            elif isinstance(result, (list, tuple)) and len(result) == len(node.outputs):
                for i, output_name in enumerate(node.outputs):
                    self.state[output_name] = result[i]
       
        print("Execution completed!")
        return self.state


def run_example(question="What are the key benefits of using a graph-based architecture for AI workflows?"):
    """Run an example LangGraph flow with a predefined question"""
    print(f"Running example with question: '{question}'")
   
    graph = LangGraph()
   
    def question_provider(state, **kwargs):
        return question
   
    graph.transform_node(
        name="question_provider",
        transform_fn=question_provider,
        outputs=["user_question"]
    )
   
    graph.claude_node(
        name="question_answerer",
        prompt_template="Answer this question clearly and concisely: {user_question}",
        inputs=["user_question"],
        outputs=["answer"],
        system_prompt="You are a helpful AI assistant."
    )
   
    graph.claude_node(
        name="answer_analyzer",
        prompt_template="Analyze if this answer addresses the question well: Question: {user_question}\nAnswer: {answer}",
        inputs=["user_question", "answer"],
        outputs=["analysis"],
        system_prompt="You are a critical evaluator. Be brief but thorough."
    )
   
    graph.visualize()
   
    result = graph.execute()
   
    print("\n" + "="*50)
    print("EXECUTION RESULTS:")
    print("="*50)
    print(f"\n
                            
                                Read More