Databricks wants to launch an internal support copilot for field engineers and account teams before the next customer conference in 10 weeks. The product is a RAG-based assistant built on the Databricks Agent Framework, using Databricks Vector Search over product docs, runbooks, release notes, and resolved support cases stored in Delta Lake and governed by Databricks Unity Catalog.
The delivery team has 9 people: 4 AI engineers, 2 data engineers, 1 PM, 1 designer, and 1 solutions architect. Leadership expects a production pilot for 600 internal users, with answers generated through Databricks Foundation Model APIs and fallback experiments on DBRX through Databricks Model Serving. The project matters because support deflection and faster field response are tied to Q4 efficiency targets.
The VP of Support wants broad coverage across all product lines at launch. The Security and Governance lead requires strict Unity Catalog access controls and no leakage of customer-sensitive case data. The Engineering Manager wants the team to avoid building custom infrastructure outside the Databricks stack. Sales Enablement wants a polished demo for the conference, even if some lower-volume use cases are deferred.
You have a fixed 10-week timeline, a remaining budget of $180,000, and no approval for additional headcount. The retrieval corpus includes 14 million support case comments, 220,000 product documentation pages, and 8,500 runbooks. A dependency team can only deliver the cleaned case-data pipeline in Week 4, and legal review of case-data usage takes 7 business days.