Inspiration
Every semester, 40,000 Terps open PlanetTerp in one tab, RateMyProfessor in another, Reddit in a third, and Testudo in a fourth, then spend hours cross referencing grade distributions, reading conflicting reviews and hoping nothing conflicts. After all that, they walk into class on day one knowing nobody. We thought: what if one conversation could replace all of that?
What it does
Shell is a voice first AI academic advisor that talks to you, learns your goals and builds your perfect semester i.e. the right courses, the right professors and the right people. You talk to Shell (named after Testudo's shell) like you'd talk to a smart upperclassman. It figures out if you're trying to get into a research lab, coast to protect your GPA for a job offer, or explore different majors then asks tailored follow ups. Along the way, it naturally learns about you where you're from, your hobbies, your clubs which powers the social matching.
Hit Next, and Shell pulls real data from PlanetTerp, Google Scholar and Testudo to score and rank courses for your specific goal. A research student sees lab openings and publication history. A coasting senior sees GPA simulations and workload stats. An exploring sophomore sees which majors took which courses and where alumni ended up.
Then Your Circle matches you with classmates who share your interests, complete with conversation starters so you never have to cold open a stranger.
On the university side, an admin dashboard gives department chairs demand forecasting, major migration signals, at risk student detection, and institutional nudges soft weights that steer recommendations toward university priorities.
How we built it
React + Vite for both the student app and admin dashboard. Kimi K2.5 (an open source LLM) via Ollama's free cloud endpoint as the conversational AI brain, with a carefully engineered system prompt that keeps conversations natural (1-2 questions per turn, adapts by persona, outputs structured JSON for the backend). Thanks to ElevenLabs we had a conversational AI agent that made the student onboarding a natural smooth process.
Classic Python fast API back end with integration with the PlanetTerp API for grade distributions and professor reviews and the Jupiterp API for course information. A scoring engine with weights that shift per goal mode i.e. research weights lab activity, coast weights %A rate, explore weights teaching quality and major diversity. The admin dashboard features animated KPI counters, hover reveal bar charts, click to drill donut charts, before/after nudge impact visualisations and a RAG powered chat sidebar.
Note: We did try and found that Terp AI can be integrated in multiple ways and aspect to the project but given the short duration of time and some other constraints we have included complete ai agentic workflow integration with it in a future roadmap.
Challenges we ran into
Making AI feel like a conversation, not a form. Getting Shell to ask one question at a time, react naturally, skip what it can infer, and transition from academic questions to personal ones without feeling like a survey took dozens of prompt iterations.
Figure out the match making algorithm and see how would that pan out with the various other aspects in the project as agentic workflows took some back and forth and debugging.
Trying to integrate Terp AI as an API through a work around had its own set of challenges such as bearer token timing out and SSL certification issues that we had to find fixes to.
Accomplishments that we're proud of
Three completely different course card UIs that adapt to your persona: research students see lab openings and Google Scholar papers, coasting students see GPA impact simulations and workload breakdowns, exploring students see alumni career paths and major distribution charts. Same product, three different experiences depending on who you are.
The university admin dashboard with institutional nudges and sliders that let UMD soft weight recommendations (boost research, fill TA pipeline, support retention) with live before/after impact visualisations proving ROI. This is how Shell becomes a business, not just a tool.
What we learned
The biggest insight: the data students voluntarily give a tool they love is worth more than any survey. Every Shell conversation reveals intent what students want, whether they're switching majors, if they're struggling at a resolution no institutional survey can match. The product is the student tool. The business is the data layer.
What's next for Shell
Course registration happens twice a year. We want Shell to bring you back every day: a personalised weekly feed (TerpLink events, deadlines, office hours), ongoing study group management within Circle matches, a research radar tracking lab openings and conference deadlines, scholarship and TA position alerts and alumni career path suggestions. The semester planner is the hook. The daily companion is the product.
Log in or sign up for Devpost to join the conversation.