A Generative Al Engineer is helping a cinema extend its website's chat bot to be able to respond to questions about specific showtimes for movies…

Databricks Feature Serving is the optimal solution for this use case. It is specifically designed to provide low-latency, real-time access to features (like movie showtimes) stored in Delta tables for applications like chatbots. By defining a FeatureSpec and creating a serving endpoint, the system automatically syncs the data to a high-performance online store. This endpoint can then be easily integrated as a tool within the RAG agent's logic. This approach is the most performant due to the optimized online store lookup and requires the least effort as it leverages a managed, purpose-built service within the Databricks platform, avoiding complex custom implementations or external infrastructure.

B: Using a text-to-SQL LLM adds unnecessary complexity, latency from the LLM inference, and potential security risks. It is less performant and more effort than a direct API call.

C: Vector search is designed for semantic similarity on unstructured text, not for precise, filtered lookups of structured data like showtimes. This is a misapplication of the technology.

D: This option introduces external infrastructure (MySQL) and the overhead of managing a separate database and a data synchronization pipeline, which is significantly more effort than using the integrated Feature Serving.

---

1. Databricks Official Documentation, "What is Databricks Feature Serving?": This document states, "Databricks Feature Serving makes features in the Databricks platform available to models or applications deployed outside of Databricks... providing a high-availability, low-latency service for serving features." It explicitly mentions RAG applications as a primary use case: "For RAG applications, the serving endpoint can retrieve features from a vector store." While it mentions vector stores, it also serves from any Delta table, and its core purpose is low-latency serving, which is key to the question.

2. Databricks Official Documentation, "Create feature serving endpoints": This guide details the process of creating an endpoint using a FeatureSpec. The simplicity of the FeatureSpec configuration, which directly references the source Delta table and primary keys, supports the "least effort" claim. The documentation notes its suitability for "online applications that require low latency."

3. Databricks Official Documentation, "Agent framework on Databricks": This page describes how RAG agents can be equipped with tools to interact with external systems. It states, "Tools are functions that an agent can learn to use... For example, a tool can query a knowledge base, a database, or any other API." A Feature Serving endpoint provides exactly this kind of API for the agent to query structured data, fitting perfectly into the described architecture.