Question 6

Explanation

The challenge is to build a specialized agent that can securely access both a private, static knowledge base (internal documents) and dynamic, real-time data (interest rate API). The solution must be efficient to build and maintain.

INCORRECT: Buying a generic, third-party chatbot SaaS product.
A generic SaaS product would be difficult to integrate securely with the bank's internal data and proprietary APIs. It also poses a data governance and security risk, as sensitive customer data would be handled by an external vendor.

INCORRECT: Training a new foundation model from scratch on the bank's data.
Training a foundation model from scratch is a massive, multi-year undertaking requiring immense data and compute resources. It is completely impractical and cost-prohibitive for this specific business application.

INCORRECT: Fine-tuning an open-source LLM on conversation transcripts.
Fine-tuning can teach the model a conversational style or specialized knowledge but does not inherently give it the ability to access real-time information from an API or search a private knowledge base. This approach only solves part of the problem.

CORRECT: Using Vertex AI Agent Builder to ground a model in their data and connect to the API via tools.

Vertex AI Agent Builder is the purpose-built solution for this exact scenario. It provides a framework to ground a model in specific enterprise data (via Vertex AI Search) and to connect it to external systems and APIs using tools. This allows for rapid, secure development without starting from scratch.