🏆 Track Winner: AI-Enhanced Academic Advising
Inspiration
From lengthy, confusing degree audits to long lists of major requirements, staying on track to graduate can be stressful. Students aren't always sure whether they're satisfying program rules, which leads to an influx of repetitive emails for academic advisors and a lot of uncertainty for students.
We wanted to build something that would make academic advising at Tufts feel clearer and more modern—helping students and advisors navigate their academic journey with less confusion and fewer back‑and‑forth emails.
What TuftsAdvisor Does
TuftsAdvisor is a two‑part solution designed to modernize academic advising:
- Advising chatbot interface: A web chat experience that students and advisors can use to ask questions about academic policies, degree progress, and requirements, and receive quick, grounded answers powered by a RAG pipeline.
- Outlook plugin for advisors: An Office add‑in that reads incoming advising emails and uses our API plus retrieval‑ augmented generation to create high‑quality draft responses, saving advisors time while keeping tone and content consistent.
Together, they reduce repetitive work for advisors and give students faster, more reliable guidance about their path to graduation.
How We Built It
On the backend, we built a Flask API that orchestrates a retrieval‑ augmented generation (RAG) workflow with GPT‑4o‑mini. Academic rules, degree audits, and policy documents are embedded and stored in priority‑based vector stores so the system can surface the most relevant passages for each question.
We used Microsoft Graph API to detect new emails in advisor inboxes and pass them through our RAG pipeline to generate structured, outlined draft responses. The same API and backend power a React web‑based chat interface for students, built with Vite and a responsive front end.
The infrastructure runs on Azure, with data persisted in MongoDB Atlas. We implemented secure login and session handling using OAuth 2.0 (MSAL.js) so each advisor can configure their own tone, signature, and reply preferences.
Challenges We Ran Into
- New concepts and tools: Many of the techniques we used—especially production‑style RAG, Graph API integration, and Microsoft Office add‑ins—aren't covered in typical CS coursework, so we had to learn them quickly.
- Navigating a large, mature codebase: Working within the Microsoft 365 add‑in ecosystem required carefully reading documentation and validating what was technically feasible inside Outlook's constraints.
- Balancing accuracy and usability: We needed responses that were both legally/academically correct and easy for students to understand, which meant iterating on prompt design and retrieval strategies.
Impact & Accomplishments
- Led a team of six to deliver a working Outlook plugin and web chatbot within the hackathon timeline.
- Cut advisor response times by an estimated 40% by generating high‑quality email drafts directly inside Outlook.
- Integrated GPT‑4o‑mini with a dual‑layer RAG store to improve retrieval accuracy by roughly 35% compared to a naive single‑store baseline.
- Won the Track Winner prize for AI‑Enhanced Academic Advising at JumboHack 2025.
What We Learned
We learned how to build and ship a Microsoft Office add‑in end to end—from manifest configuration and sideloading to integrating with live mailboxes through Microsoft Graph. On the AI side, we deepened our understanding of how to design and evaluate RAG systems, including document chunking, priority stores, and prompt templates that reduce hallucinations.
We also gained experience in managing cloud infrastructure on Azure, wiring together React, Flask, and MongoDB Atlas with secure auth, and collaborating as a cross‑functional team under tight time constraints.
Tech Stack
- Frontend: React, Vite, CSS
- Backend: Python, Flask, REST API, MongoDB Atlas
- AI: GPT‑4o‑mini, retrieval‑augmented generation (RAG) with dual‑layer priority stores
- Integrations: Microsoft Graph API, Outlook add‑ins, MSAL.js (OAuth 2.0)
- Infrastructure & tools: Azure, Git