Enterprise RAG Challenge - Solutions

by Felix Krause

Solutions to the Enterprise RAG Challenge of Timetoact Austria on 27.02.2025.

BLOG POST explaining the solutions here.

Current Solutions

• Google Gemini 2.0 Flash - Naive Approach: full PDF(s) in context
• Multi-agent approaches (openAI based router to extend and specialise queries for each company)
  • Qdrant RAG
    • Custom chunking of markdown file obtained via docling parser
    • Then Qdrant for vector database retrieval of chunks (top 5)
    • openAI o4 or IBM granite-20b-code-instruct based answer generation per specialised company query, and o4 based final answer generation
  • Gemini Retrieval and openAI Generation
    • Gemini 2.0 Flash with full PDF in context for retrieval
    • OpenAI for final answer generation based on Gemini's company specific answers

Environment

Create environment from environment.yml file:

mamba env create -f environment.yml

Export environment.yml:

mamba env export > environment.yml

Tracing

To trace, first start a trace server:

python -m phoenix.server.main serve