Project Planning: Building the Graph Schema

Welcome to the Capstone Project. Throughout this course, we have learned the building blocks of LangGraph: state, nodes, edges, and advanced patterns like human-in-the-loop and parallel execution. Now, it is time to put them all together.

In this module, we will build a Multi-Agent Research and Coding Assistant. In this first lesson, we focus on the Blueprint: Designing the Graph Schema.

---

1. The Project Objective

Our goal is to build an agent that can:

1. Search for documentation about a new library.
2. Plan a small code example using that library.
3. Execute the code (simulated) and check for errors.
4. Refine the code if it fails.

---

2. Defining the Unified State

A complex system requires a structured state. We need to track the user's request, the research findings, the proposed code, and any error messages.

from typing import TypedDict, List, Optional

class ProjectState(TypedDict):
    task: str                 # The user's original goal
    research_notes: str       # Consolidated findings from the web
    proposed_code: str        # The logic generated by the agent
    execution_result: str     # Success/Fail message from the runner
    iteration_count: int      # To prevent infinite loops (Module 7)
    is_approved: bool         # Human-in-the-loop flag (Module 12)

---

3. Visualizing the Nodes

Before coding, we map out the "Departments" of our agent:

1. ResearchNode: Uses a search tool to gather data.
2. ArchitectNode: Takes the notes and writes the code.
3. RunnerNode: Simulates running the code and returns errors.
4. HumanGate: Pauses for user approval before final "Shipping."

---

4. The Edge Logic (The Decisions)

The "Brain" of the graph lives in the conditional edges.

• IF the code fails in the RunnerNode -> Route back to the ArchitectNode.
• IF the iteration count > 3 -> Route to a "Failure" exit.
• IF the code passes -> Route to the HumanGate.

---

5. Setting the Entry and Exit Points

• Entry: The graph always starts at ResearchNode.
• Exit: The graph terminates at a Success node or a Max_Retries fallback.

---

Summary and Mental Model

Think of this lesson as Architecting a House.

• We aren't laying bricks yet.
• We are drawing the Blueprints (The Graph) and deciding where the Plumbing (The State) goes.

A well-planned graph is 80% of the work in production AI.

---

Exercise: Schema Design

1. Expansion: If you wanted to add a "Documentation Node" that writes a README for the code, where in the graph would you put it?
2. Safety: Why is it important to include iteration_count in the ProjectState instead of just letting the graph loop forever?
3. Logic: Draw a rough diagram of this graph on a piece of paper. (Nodes, Edges, and the loop point). Ready to start the departments? Next lesson: Implementing the State and Logic Nodes.