IntermediateLangGraphTavily APIOpenAI

Build a Sales Agent

In this project, we will build a multi-step agent using LangGraph. The agent will accept a company name, research it using a search tool, identify key value propositions, and draft a cold outreach email.

Architecture

User Input (Company Name) 
  --> [Research Node] --> Search API
  --> [Synthesis Node] --> Identify Pain Points
  --> [Drafting Node] --> LLM Generation
  --> Final Email Draft

Implementation Steps

Environment Setup

Install LangGraph and configure API keys for OpenAI and Tavily.

Define the State

Create a TypedDict state to hold the company info, research notes, and draft.

Implement Research Node

Write a function that calls the search API and summarizes top 3 results.

Implement Drafting Node

Prompt the LLM to write an email based *only* on the research notes.

Connect the Graph

Wire the nodes together and compile the graph.

Code Structure

agent.py

from typing import TypedDict
from langgraph.graph import StateGraph, END

class AgentState(TypedDict):
    company: str
    research: str
    draft: str

def research_node(state: AgentState):
    # Call search tool
    return {"research": "..."}

def draft_node(state: AgentState):
    # Call LLM
    return {"draft": "..."}

workflow = StateGraph(AgentState)
workflow.add_node("research", research_node)
workflow.add_node("draft", draft_node)
workflow.set_entry_point("research")
...

Ready to start?