A company wants its employees to engage with an HR policy assistant, powered by Retrieval Augmented Generation (RAG), directly within Slack, their primary communication platform. Considering an AI agent hosted on Databricks, what is the MOST EFFECTIVE method to integrate this agent and make it available in Slack?
Answer & explanation
Correct answer: A
WHY A: The standard pattern for integrating a Databricks Model Serving endpoint with Slack is: (1) create a Slack App/Bot using Slack Bolt (Python SDK), (2) on each incoming Slack message, call the Databricks endpoint's REST API (OpenAI-compatible chat completion endpoint) with the user's message, (3) return the response to the Slack thread. The service principal token is stored securely as an environment variable in the bot's hosting environment (Databricks Apps, cloud function, or container) and never hardcoded.
WHY NOT B: Hugging Face Spaces host independent model demos and require the model to be deployed on Hugging Face infrastructure. This would mean migrating the agent out of Databricks and losing Unity Catalog governance, Vector Search integration, and Databricks authentication, and a web link is not an in-Slack experience.
WHY NOT C: A public unauthenticated endpoint exposes the HR policy assistant to arbitrary internet traffic. HR data queries must be authenticated to prevent unauthorized access to company policies.
WHY NOT D: Slack does not have a native LLM connector or plugin that can directly query Unity Catalog tables or Model Serving endpoints. Databricks endpoints are accessed via REST API, not through a Slack-native protocol.