Databricks Generative AI Engineer Associate Question 23
Select 2You are building a Generative AI application on Databricks that answers questions based on a large dataset of scientific articles. The desired workflow is: (1) Retrieve relevant articles based on the query; (2) Summarize the retrieved articles; (3) Generate a final response using the summarized content. Which combination of chain components would be most appropriate for this workflow?
- A
Retrieval-based QA chain, Summarization chain, LLM chain
- B
Retrieval-based QA chain, Embedding chain, Summarization chain
- C
Embedding chain, Summarization chain, LLM chain
- D
Vector Store Retriever chain, Summarization chain, LLM chain
- E
Vector Store Retriever chain, Embedding chain, Retrieval-based QA chain
Show answer and explanation
Correct answers: A, D
Explanation
To achieve the desired input-output workflow, you need components that retrieve relevant articles, summarize them, and generate a response. Retrieval-based QA chain or Vector Store Retriever chain can handle the retrieval step. A Summarization chain condenses the retrieved content, and the LLM chain generates the final output. Embedding chains are not necessary for this workflow, as embeddings are typically used for vectorization, not direct summarization or response generation.
- A. Correct.
Correct: This sequence aligns with the workflow. The Retrieval-based QA chain retrieves relevant articles, the Summarization chain condenses the content, and the LLM chain generates the final response.
- B. Incorrect.
Incorrect: While the Embedding chain can help encode data, it is not used directly in the workflow described. Summarization would follow retrieval, not embeddings.
- C. Incorrect.
Incorrect: Embedding chain is unnecessary in this case, as the workflow already specifies retrieval, summarization, and response generation.
- D. Correct.
Correct: A Vector Store Retriever chain retrieves relevant documents, the Summarization chain condenses them, and the LLM chain synthesizes the final response.
- E. Incorrect.
Incorrect: While Vector Store Retriever chain is relevant for document retrieval, Embedding and Retrieval-based QA chains are redundant or do not fit the summarization and response generation steps.