Vanguard - “Student Concierge” Agent

I am 99% sure this is going to run, but am awaiting final confirmation. You may claim it in the meanwhile.
Build a Drexel-focused AI concierge that answers student questions using grounded retrieval (RAG) from trusted sources and can optionally take actions (e.g., create reminders, draft emails, generate checklists) with strong prompt-injection defenses.

Problem statement

Students lose time navigating scattered information (deadlines, forms, policies, program requirements). A concierge that only answers from validated sources and exposes “why/how I know this” to meaningfully improve student experience.

Target users & sample use cases

First-years: “How do I find advising / tutoring resources?”
Co-op students: “What’s my checklist for next term?” (generalized; you can map to Drexel’s experiential model)
Seniors: “What are my key deliverables for my senior capstone project?”
Core features (MVP scope)

Curated knowledge base from approved public sources (web pages, PDFs, handbooks) + provenance display (“Answer grounded in: …”).
RAG pipeline with source whitelisting, chunking, retrieval, and citations. (Implementation details are up to the team.)
Safety & security layer: prompt-injection detection + content sanitization + “confirm before action.”
Human-in-the-loop mode: “Draft” vs “Send” for emails/messages; user confirmation required for any tool use.
Stretch goals

“Ask the Knowledgebase” and “Do the thing”: create calendar reminders, generate a weekly plan, or fill out a checklist. (Tooling can be mock integrations if needed.)
Add evaluation harness (accuracy, citation correctness, jailbreak/prompt-injection robustness).
Deliverables

Web app (chat UI) + backend service + indexed Knowledgebase
Security report: threat model + mitigations (prompt injection focus)
Evaluation report: grounded-answer precision, citation fidelity, and adversarial tests