knowledge-retrieval-api

Standalone Knowledge Retrieval API Server to be used with Rubra

Development

• Run in development mode (hot-reloading): make run-dev (Requires docker and compose)
• Dependency Management: uv
• Linting & Formatting: ruff

File Types

Currently, the following file types are supported for ingestion via llama-index' SimpleDirectoryReader interface:

• .csv - comma-separated values
• .docx - Microsoft Word
• .epub - EPUB ebook format
• .hwp - Hangul Word Processor
• .ipynb - Jupyter Notebook
• .jpeg, .jpg - JPEG image
• .mbox - MBOX email archive
• .md - Markdown
• .mp3, .mp4 - audio and video
• .pdf - Portable Document Format
• .png - Portable Network Graphics
• .ppt, .pptm, .pptx - Microsoft PowerPoint

Examples

You can use the GPTScript example in the examples/ directory to test the ingestion and querying parts of the API. The GPTScript will do the following:

1. Ingest the llama2 Paper located as examples/data/llama2.pdf (only if it hasn't been ingested before)
2. Query the Dataset to tell us something about the topics "Truthfulness, Toxicity, and Bias"

The returned response should contain a reference to the source page.

Just run this from the repository root:

# Optional: Create a Python venv

# Install the requirements
pip install -r requirements.txt

# Run the server
make run-dev # if you haven't already

# Create the dataset
curl -X 'POST' \
  'http://localhost:8000/datasets/create' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "name": "llama2",
  "embed_dim": 0
}'

# Run the GPTScript example
gptscript examples/example.gpt