LangGraph Orchestrated RAG Over Nike 10-K

📘 Description

This project demonstrates a LangGraph-orchestrated Retrieval-Augmented Generation (RAG) system for intelligent querying of Nike’s 2023 10-K filing. It extends prior modules by introducing a graph-based control flow, allowing the app to analyze questions, retrieve relevant vector data using metadata filters, and generate structured responses — all as composable LangGraph nodes.

This approach reflects a production-ready design for enterprise AI, combining semantic search, structured reasoning, and orchestration to deliver scalable, transparent question-answering pipelines.

⸻

🔧 Features

• LangGraph Orchestration: Constructs a structured execution flow using LangGraph, where each step (query analysis → retrieval → generation) is defined as a node.
• Mermaid Diagram Compatibility: Outputs the graph as a Mermaid diagram for inspection and transparency.
• Checkpointed Memory: Supports state management between nodes with LangGraph’s MemorySaver for persistence and reproducibility.
• Metadata-Filtered Search: Performs section-specific retrievals (e.g., only from “middle” of the document).
• Modular Design: Each RAG component — from question parsing to chunk filtering to response generation — is encapsulated as a separate function node.
• Prompt Customization: Uses minimal, safety-conscious RAG prompts for controlled output.

⸻

💡 Key Insight

Orchestration is the secret to reliable AI systems. LangGraph makes it possible to coordinate retrieval, reasoning, and memory across steps — turning one-off queries into robust workflows and enabling LLMs to act as structured agents in dynamic environments.

🔗 View the source code on GitHub