Markdown Query Engine

Provides a sample LlamaIndex application that indexes markdown files and allows for semantic search queries. Simply provide a path to the markdown files you want to index.

Prerequisites

• Python 3.10+
• Ollama
• Docker

Setup

1. Copy the .env.example file to .env and supply the namespace and data file path. Optionally, you can set OPENAI_API_KEY if you want to use ChatGPT instead of local Ollama.
2. Run pip install -r requirements.txt to install the dependencies.

Running

1. Run docker compose up to start the Redis and OpenSearch containers.
2. Run python -m MarkdownQueryEngine.bin.index to index the documents.
3. Run python -m MarkdownQueryEngine.app.main to start the FastAPI server.

Querying

Send a POST request to http://localhost:9898/query with a JSON body containing the query string.

{
    "query": "Why did the chicken cross the road?"
}

Architecture

                             User Query
                                 ↓
              "Summarize X, or give me details about Y."
                                 ↓
                        +------------------+
                        | RouterQueryEngine |
                        +------------------+
                                 ↓
                        +------------------+
                        |   LLM Selector   |
                        | (Decides Route)  |
                        +------------------+
                                 ↓
                    +--------------------------+
                    |                          |
                    ↓                          ↓
        +--------------------+      +--------------------+
        |   List Engine      |      |   Vector Engine   |
        | (Summary Store)    |      |  (Specific Info)  |
        +--------------------+      +--------------------+
        | "Summarize all    |      | "Find details     |
        |  documents..."    |      |  about X..."      |
        +--------------------+      +--------------------+
                    ↓                          ↓
                    +---------------------------+
                                 ↓
                          Final Response