Nature Finance RAG client for working with IMF climate development reports

This is a RAG application for IMF climate development reports. It is built with Next.js, Shadcn UI, and Vercel AI SDK, with a Postgres vector database to store embeddings and retrieve relevant content.

Getting Started

Follow the setup steps in the RAG API repository to set up the database and ingest the data before continuing.

Chatting with the RAG application

To chat with the RAG application, run the following command:

npm run dev

Tech Stack

Here's a breakdown of the tech stack, along with Mermaid diagrams representing the ingestion/database and client logic.

Overall Tech Stack

• Framework: Next.js
• UI Library: Shadcn UI, Radix UI React
• Language: TypeScript
• Database: PostgreSQL with pgvector extension
• ORM: Drizzle ORM
• AI/Embeddings: OpenAI embeddings (using @ai-sdk/openai and ai libraries)
• PDF Processing: pdf-parse
• Environment Variables: @t3-oss/env-nextjs

graph TB
    subgraph "Browser"
        A["Chat Client"] --> B["User Input"]
    end

    subgraph "Application Server"
        C["Chat Application Server"] --> D["LLM Service<br>(GPT-4o/Claude)"]
        D -->|"Returns messages<br>or tool calls"| C
        C -->|"If tool call"| E["Query Embedding"]
        E --> F["Vector Search"]
    end

    B --> C
    C --> A

    subgraph "Database"
        F --> G["PostgreSQL with pgvector"]
        G --> F
        F --> C
        G -- Stores --> H["Resources Table"]
        G -- Stores --> I["Embeddings Table"]
        G -- Stores --> J["Key Concepts Table"]
    end

    style G fill:#f9f,stroke:#333,stroke-width:2px

Loading

1. Ingestion/Database Logic

• Ingestion Script (ingest-pdfs.ts): The entry point for the data ingestion process. It takes a file path or directory path as input.
• PDF Processor (pdf-processor.ts): Handles the extraction of text from PDFs using pdf-parse, chunks the text, generates embeddings, and stores the text and embeddings in the database.
• pdf-parse: Used to parse the pdfs.
• Chunking Logic: Logic within the PDF Processor to split text into manageable chunks.
• @ai-sdk/openai (embedding generation): Uses OpenAI's API to generate embeddings for the text chunks.
• Drizzle ORM: An ORM used to interact with the PostgreSQL database.
• PostgreSQL with pgvector: PostgreSQL database with the pgvector extension enabled to store and perform similarity searches on embeddings.
• Resources Table: Stores the original text content of the PDFs.
• Embeddings Table: Stores the text chunks and their corresponding embeddings.

2. Client Logic

• Client (app/page.tsx): This is the main React component for the chat interface. It uses the @ai-sdk/react library.
• @ai-sdk/react (useChat hook): A hook that manages the chat state, including messages, input, and handling user input.
• User Input: The text entered by the user in the chat interface.
• app/api/chat/route.ts: A Next.js API route that handles the chat logic.
• @ai-sdk/openai (gpt-4o): Uses OpenAI's API to generate responses based on the user's input and the relevant content retrieved from the database.
• Tools (addResource, getInformation): Functions that the AI can use to interact with the system, the ai sdk provides methods to use these tools.
• addResource (actions/resources.ts): Adds a resource, called by the ai's tools to add pieces of knowledge to the RAG database.
• findRelevantContent (ai/embedding.ts): Retrieves the most relevant content from the database based on the user's query using cosine similarity.
• Drizzle ORM: An ORM used to interact with the PostgreSQL database.
• PostgreSQL with pgvector: PostgreSQL database with the pgvector extension enabled to store and perform similarity searches on embeddings. This is where the vector similarity search happens.
• Embeddings Table: Stores the text chunks and their corresponding embeddings. findRelevantContent queries this table.
• Relevant Content: The chunks that are most similar to the user's query.