A city wants residents to use a permitting pre-check chat before filing. Which Foundry generative step should developers implement first for that experience?

Deploying and consuming an LLM in Foundry for the permitting pre-check chat is the first generative step that matches the city's resident experience goal. A pre-check chat must interpret natural-language questions about requirements, documents, and eligibility hints, then generate helpful conversational replies. That pattern starts with a deployed language model in Microsoft Foundry and an application path that authenticates and calls it—not with a Foundry vector index over permit codes for the permitting pre-check chat that cannot generate replies by itself. Building a Foundry vector index over permit codes for the permitting pre-check chat supports retrieval but does not deliver a chat pre-check without a deployed language model to generate responses. Designing a Prompt Flow chat workflow with permit prompts for the pre-check chat application organizes prompts and conversation logic but does not deliver the interactive generative assistance residents need until a model is deployed and consumed. Configuring Foundry content filters for permit prompts in the permitting pre-check chat provides safety guardrails for a running system but does not create the real-time conversational pre-check until the LLM is deployed and consumed. Exam caveat: Foundry generative solutions begin with deploy-and-consume LLM patterns; later RAG, tools, and evaluations build on that foundation. Operational check: confirm the permitting app configuration points at a Foundry deployment endpoint, can obtain a token, and returns generative responses for representative pre-check prompts before layering retrieval or workflow orchestration.