Applying to universities abroad is a research project in its own right. The information every applicant needs - entry requirements, realistic total costs, deadlines, document checklists - is scattered across university websites in inconsistent formats, and each student reassembles it from scratch in spreadsheets and browser tabs. The essays, meanwhile, get redrafted with no structured feedback at all.
AICE (AI College Exploration) is a multi-agent application, built with two teammates, that takes over the structured parts of that work - collecting, comparing, and planning - so the student's time goes into the decisions only they can make.
Why agents, not one chatbot
A single general-purpose chat prompt can answer any of these questions, but its answers arrive as prose - unstructured, inconsistent between universities, and hard to double-check. We instead scoped each feature to its own CrewAI flow: an agent with narrow instructions and a defined output format. That is what lets the UI render costs as comparable breakdowns, deadlines as timelines, and requirements as checklists - and lets a user trace where each answer came from, rather than trusting one long chat response.
What it does
- AI Essay Writing Assistant - Helps students draft and refine personal statements and essays.
- University Information & Comparison - Collects and compares key university data and requirements, with automated scraping and content filtering.
- Cost Breakdown Analysis - Breaks down tuition, living costs, and total expenses - including the hidden ones - per program.
- Timeline Planner - Builds personalized application timelines with deadline reminders and buffer-time suggestions.
- Application Checklist Generator - Creates checklists of required documents and tasks for each university.
- Interview Preparation - Provides sample interview questions and preparation tips.
Short walkthrough of the AICE UI and agent flows
Architecture
A FastAPI backend orchestrates a separate run for each flow, while the Streamlit frontend provides multi-page navigation and forms. Because agent tasks are long-running, the UI polls live status rather than blocking on a response, and a lightweight SQLite database keeps session results and internal state so a student's research survives between visits. The repository follows a modular structure for agents, tasks, and utilities with environment-based configuration.
Contributors
Source, docs, and ongoing updates are available on the GitHub repository.