The Student Success Navigator team developed a conversational AI triage assistant called Sammy for San Jose State that routes students to campus resources by interpreting natural-language questions and connecting them to tutoring, financial aid, career services, EOP, and accessibility support offices with grounded guidance and citation cards linked to official SJSU pages. The solution is built using an Astro and React frontend with a FastAPI backend that implements a safety-first architecture where crisis keywords like suicide, self-harm, or assault are matched deterministically before any AI processing, triggering an immediate handoff with 988 Lifeline, CAPS, and emergency contacts without model improvisation. Campus resource information is stored as Markdown documents with metadata sidecars in Amazon S3 (cloud storage), which feeds into Amazon Bedrock Knowledge Bases (managed RAG) that uses Amazon Titan Text Embeddings to create vector representations stored in Amazon OpenSearch Serverless (search and vector index), enabling semantic search across 800+ official SJSU pages. When a student asks a question that passes the safety gate, the backend calls the Amazon Bedrock (managed generative AI) Converse API with Claude Sonnet 4.6 models in an agent loop where the model decides whether to retrieve campus resources by calling a retrieve tool that queries the Knowledge Base or to respond conversationally without retrieval, and when retrieval occurs, the system shapes the returned chunks into deduplicated citation cards showing the source URL, page title, and relevant excerpts. The agent submits its final answer through a structured tool call that bundles conversational text with statement batches and resource cards, ensuring responses are grounded in exact text from official pages with citations, and the system includes Amazon Cognito (authentication) infrastructure for Google OAuth with @sjsu.edu pre-sign-up validation, though authentication is not yet connected to the chat interface. This architecture provides students a low-barrier self-service channel that offloads routine navigation questions from staff, reduces office-to-office referral chains by routing directly to the correct service, and improves resource discoverability for students hesitant to ask for help in person, with the deterministic safety gate ensuring crisis situations bypass AI entirely for immediate human support.