What problem does it solve? Financial-services GenAI demos often produce ungrounded chatbot output with no citations, audit trail, or human oversight, making them unusable for regulated fraud and banking workflows. This Skill guides the construction of Layer 3 of a Databricks FSI solution: governed retrieval, grounded generation, evaluation, and audit evidence. ## Core Features & Use Cases - Grounded Case Summaries: Join trace IDs to customer history, predictive model scores, and versioned policy documents to generate cited case summaries with evidence, counter-evidence, and recommended actions. - Governance and Guardrails: Enforce caller data permissions, treat retrieved text as untrusted, restrict actions, and require analyst ownership of final decisions. - Evaluation and Audit: Build versioned evaluation sets covering groundedness, citation correctness, safety, privacy, latency, and cost, and persist prompt/model versions with trace IDs for reproducibility. - Use Case: A fraud investigator copilot that summarizes a flagged case, cites the exact policy and resolved-case sources, recommends a next action, and abstains when evidence is insufficient. ## Quick Start Ask the agent to build a grounded fraud case summary workflow on Databricks that retrieves governed policy documents, generates a cited recommendation, and records audit evidence by trace ID.
