Academic Advising Case Study

Inspiration

Every TCU Neeley student plans four years of courses from documents that don't agree with each other: major checklists, the undergraduate catalog, the course list, the registrar's degree audit and AP credit tables. Mistakes usually surface late, at registration or in senior year, when fixing them costs a whole semester.

When we dug in, it was worse than we expected. On the Finance checklist, 4 of 13 listed electives (31%) can't actually be taken by regular Finance majors: one is restricted to Real Estate concentration students, and the other three require it.

Using TCU's published 2026–27 cost of attendance, one avoidable extra semester costs a student:

One extra semester = $44,871

Tuition: $66,520 ÷ 2 = $33,260
Housing, food, books, travel and personal costs: $23,132 ÷ 2 = $11,566
Student fee: $90 ÷ 2 = $45

We wanted a tool that catches these problems before a student does.

What it does

Graduation Blueprint turns a student's record into a checked, semester-by-semester plan to graduation.

Three ways to start: upload a TCU Academic Advising Report (read in the browser; name and student ID are discarded), start as a first-year with no report yet, or describe yourself in plain words.
A real plan: it covers all 8 Neeley majors, 8 minors and 3 certificates, with prerequisites, C- minimums, fall-only and spring-only courses, hour caps, catalog-year rules and AP/IB/CLEP credit.
Suggestions and what-ifs: ways to finish sooner, plus side-by-side comparisons for switching majors or adding a minor.
Find your major: an 8-question survey ranks all 8 Neeley majors and explains each match.
Get involved: 43 Neeley programs, centers and clubs, matched to the student's major, interests and year.
Advisor handoff: anything the tool can't verify becomes a list of exact questions for the student's advisor.

How we built it

Our design rule was rules decide, AI explains.

Data: we collected requirements from 16 official TCU sources and turned the team's workbooks (3,888 courses, 22 programs, 126 exam-credit rows) into JSON with Python importers. Every rule carries its source ID.
Rules engine: a JavaScript engine builds each student's model of what's done, in progress and still needed. It then schedules the remaining courses term by term, respecting prerequisites, offering terms and hour caps.
Independent validator: a separately written checker re-tests every finished plan. A plan that fails is labeled a draft and sent to an advisor.
AI layer: Claude does two jobs. It turns a student's free-text description into proposed changes, and it explains finished plans in plain English with source citations. Every course code Claude returns is checked against the catalog, and nothing changes until the student presses Apply.
Testing: we wrote 47 engine checks in Node, plus 17 in-app test scenarios (82 checks) that judges can run live. Playwright browser tests cover desktop and phone layouts.

The whole app ships as a single HTML file: no server, no database, no API keys.

Challenges we ran into

Sources that disagree: the Strategy course is MANA 30350 on some checklists and MANA 40250 on others, and checklist prerequisites are often looser than the official course descriptions. We recorded 15 known conflicts, and the engine applies the stricter rule wherever two sources disagree.
When the AI got it wrong: we used AI to pull prerequisites out of thousands of course descriptions, and it read phrases like "or concurrent enrollment" and "may not receive credit for both" as required courses. Our own rule checks caught it when plans for some majors started breaking. We wrote explicit rules for those phrases and labeled all 105 machine-read prerequisite sets "auto-read" so advisors know which rules weren't hand-checked.
Knowing when to say no: for a transfer student with unmapped credit on an older catalog, the right answer is to refuse to certify the plan and send the student to an advisor, not to guess.
First-year edge cases: our first freshman plans showed students finishing a semester late, because the semester in progress counted as zero hours. We now assume a full 15-hour first term unless the student lists courses, and the screen says so.
Privacy: degree audits contain real student data. The PDF is parsed entirely in the browser, identity fields are discarded, and every demo student is made up.

Accomplishments that we're proud of

A working tool, not a mockup: it plans for any student profile across 19 Neeley programs, and every one produces a plan that breaks no rule.
Every check passes: 47 of 47 engine checks, and 17 of 17 test scenarios (82 of 82 checks) running live on the app's Validation page.
Ten findings in TCU's own data, each tied to an official source, including the sharpest bottleneck: Finance electives closed to Finance majors.
Responsible AI by design: the language model can't change a requirement, every AI-suggested course code is verified, and the tool knows when to send a student to a human.
Built for first-years too: students can get a full four-year plan, a major match and club recommendations before they ever have a degree audit.

What we learned

Deterministic rules plus AI explanation beats AI alone for anything with right and wrong answers. The language model is excellent at reading messy input and explaining results, but it should never decide a graduation requirement.
Validation has to be independent: most of our bugs were caught by the checker that shares no code with the planner.
The data problem is the real problem: most of our findings, like the closed electives and the two Strategy numbers, are data issues TCU can fix at the source.

What's next for Academic Advising Case Study

We propose a 12-month pilot with the Neeley Academic Advising Center and the Registrar:

Months 1–3: connect a read-only registrar feed for real section times, and complete privacy and accessibility reviews.
Months 4–6: an advisor-led beta in about 50 advising appointments.
Months 7–9: an opt-in pilot for 500 first-year and sophomore Neeley students.
Months 10–12: measure results and decide on expanding to all Neeley majors.

If 5% of pilot students would otherwise add a semester and the tool prevents 30% of those delays: Pilot impact ≈ $336,533 in student costs avoided