Senior Capstone · St. Edward's University

Course advising,
on demand

An AI-powered chatbot trained on St. Edward's University course data — giving students accurate, personalized course recommendations without waiting for office hours.

TypeSenior Capstone
RoleUX · Frontend · API · GPT Training
TeamCollaborative (partners: Python/Flask)
StackOpenAI GPT · Python · Flask · Figma
Watch Demo ↗ Read Case Study ↓
scroll
01 — Overview

The Problem

Course registration at a university is frustrating. Students don't always know what to take, when to take it, or how it fits their degree plan. Academic advisors are overloaded. The information exists — it's just buried. Topper was designed to bridge that gap.

⚠ The Gap
Students wait days for advisor meetings and still leave confused about electives
The advising bottleneck means students make uninformed decisions about their schedules, or delay registration entirely.
✓ The Fix
AI chatbot trained on real course data gives answers in under 5 minutes
Topper provides instant, personalized course recommendations based on actual university data — no appointment needed.
02 — My Role

What I contributed

This was a collaborative senior capstone project.

🔍
UX Research
Understanding how students navigate course selection; where the pain points were.
User Research
🎨
Frontend Design
Figma prototype for the full chat interface and user flow.
Figma
🔗
API Configuration
Integrated the OpenAI API and set up the data pipeline.
OpenAI API
🤖
Custom GPT Training
Built 500+ unique training queries to make the model give accurate, university-specific answers.
500+ Queries
🛠
Infrastructure Pivot
When original plan fell through, led the pivot from Google Cloud to updated OpenAI API platform mid-project.
Problem Solving
03 — Process

How we built it

Step 1
User Research
Identified that students needed fast, specific answers about courses, prerequisites, and degree requirements.
Step 2
Scoped the AI
Determined what the chatbot should and shouldn't know — setting clear boundaries for accurate, useful responses.
Step 3
Built training dataset
Created 500+ queries covering courses, prerequisites, degree requirements, and scheduling to train the model on university-specific knowledge.
Step 4
Designed the interface
Conversational UI designed in Figma for both mobile and desktop — focused on clarity and speed.
Step 5
User Testing (2 rounds)
Tested whether students could find a relevant course in under 5 minutes. Result: yes — across both rounds of testing.
04 — Results

What we achieved

<5 min
To find relevant courses (both testing rounds)
500+
Custom training queries
2
Rounds of user testing
1
Major infrastructure pivot navigated
05 — Reflection

Lessons learned

Topper was built and tested but not publicly released.

"Custom training data is the difference between a chatbot and a useful one — generic prompts don't cut it for specialized knowledge."
— Andre Espinoza
📚
Training data is everything
Generic prompts don't cut it for specialized knowledge. The 500+ custom queries made the difference.
🔄
Pivots are design problems
Mid-project infrastructure pivots require design thinking, not just technical fixes.
🤝
Teaching is building
Teaching non-technical teammates how to use AI tools was as valuable as the build itself.

See more of my work

Topper AI was built and tested as a senior capstone — explore the demo or head back to the portfolio.

Watch Demo ↗ ← Back to Portfolio