Code Section

Scenario Section

Task Description: Reduce LLM hallucination via Retrieval-Augmented Generation (RAG).

  1. Construct a ‘Category-Product’ tree.

    All Products - Support and Downloads

    graph.png

  2. Retrieve top-1000 relevant documents for each product.

    Untitled

  3. Design an representation for searching.

    Input: the question from customers

    Output: The top-k relevant documents

    References: Graph Representation Learning / Dense Vector Search / Two-tower Architecture

  4. Design a mechanism to utilize the retrieved documents in LLM generations (optional)

    Context Windows / References / Evaluation