Implement initial version Retrieval-Augmented Generation (RAG) System for Kyma Agent

[Teneroy]

Description

The goal of this task is to implement a Retrieval-Augmented Generation (RAG) system for the Kyma agent. The RAG system will enhance the Kyma agent's ability to provide accurate and relevant responses by retrieving relevant documents from a knowledge base and combining the retrieved information with a generative model. The system should include proper query-translation techniques, efficient information retrieval, and a reranking mechanism to improve the quality of retrieved information.

Ensure that the flow works, necessary steps are implemented. Further optimization of performance will go in the following tickets

Subtasks

1. Query Translation:

   • Implement query-translation techniques to transform user queries into an optimal format for the retrieval system.
   • Ensure that the Kyma agent can handle diverse and complex queries by translating them into effective search queries.
   • Test different query-translation strategies to enhance the accuracy and relevance of the retrieval process.

2. Information Retrieval:

   • Develop a retrieval mechanism to pull relevant documents or data from the knowledge base, such as Kyma documentation or related resources.
   • Integrate the Hana Vector Database client as the main source for retrieving embedded information.

3. Reranking Mechanism:

   • Implement a reranking system that scores and ranks the retrieved documents based on their relevance to the original query.
   • Use relevant metrics or machine learning-based approaches to ensure that the most important and accurate information is ranked higher.
   • Test the reranking system with a variety of test cases to validate its effectiveness.

Acceptance Criteria

• Query-translation techniques are implemented and improve the quality of search queries - Mansur
• The information retrieval system pulls relevant documents from the knowledge base efficiently
• A reranking mechanism is implemented and significantly improves the relevance of retrieved documents.