The Ask Eddie team developed an AI-powered student assistant chatbot that provides 24/7 support for Lemoore College, addressing the problem that 50% of calls went unanswered during the first week of school, which previously led to students not enrolling, high dropout rates, and lower overall satisfaction. Built as a Next.js application hosted on AWS Amplify (full-stack hosting), the system routes every student question through a deterministic validation pipeline that begins with Amazon Bedrock Guardrails (content filtering) screening for sensitive information before proceeding to retrieval, ensuring that requests containing personally identifiable or prohibited content are rejected immediately rather than processed. Approved questions are sent to Amazon Bedrock Knowledge Bases (managed RAG), which indexes curated college documents stored in Amazon S3 (cloud storage) and uses Amazon OpenSearch Serverless (search & vector index) as the underlying vector search engine to retrieve relevant passages grounded in official Lemoore College sources such as admissions policies, financial aid FAQs, academic calendars, course dates, and office contact information. The system calls Amazon Bedrock (managed generative AI) foundation models to compose answers with citations from the retrieved passages, applies deterministic escalation rules to decide whether the query requires human handoff based on confidence scores and evidence sufficiency, and returns a typed response that clearly indicates whether the answer came from the knowledge base or should be escalated to front-desk staff. Student feedback and minimized, redacted analytics are stored in Amazon DynamoDB (NoSQL database) without retaining raw sensitive content, and Amazon Cognito (authentication) provides role-based access control for ambassador and admin dashboards where staff can review unanswered questions, sync approved document sources, and monitor system health. The team implemented comprehensive trust and safety measures including a curated test set with a grading rubric that checks answer accuracy and defends against OWASP Top 10 LLM risks such as prompt injection and excessive agency, security-by-design principles with robust testing documentation built into the repository, and human escalation paths that route complex or sensitive queries to live staff rather than forcing the AI to answer beyond its capability. As a stretch feature, the team demonstrated a voice call interface prototype that allows students to speak their questions and receive spoken answers, extending the chatbot’s accessibility to students who prefer calling over typing. The solution provides students with instant answers at any hour, reduces the burden on overloaded front-desk staff who can now focus on complex cases, improves access for potential students exploring the college, and lays a foundation for future enhancements such as full 24/7 voice call support, automated continuous testing, and hardened production security frameworks.