Academic Copilot

Inspiration

Scholarships can make a huge difference for students, but finding the right ones is often unnecessarily difficult. Students have to search through long scholarship directories, open individual award pages, compare eligibility requirements, keep track of deadlines, and repeatedly enter information that already exists in their academic record.

We wanted to reduce that process from hours of manual searching into a guided workflow. The idea behind Academic Copilot was to create an AI agent that already understands a student's academic profile and can actively help them discover opportunities instead of requiring them to search for everything themselves.

What it does

Academic Copilot connects to a student's university academic record and turns that information into a personalized scholarship and academic dashboard.

It can:

retrieve the student's GPA, completed credits, program, year of study, and course history
calculate academic performance and scholarship history using deterministic logic
search official university scholarship sources
rank scholarships based on the student's actual academic profile
explain why a student may or may not qualify
identify missing eligibility information and ask the student targeted follow-up questions
track scholarship deadlines and link back to official sources
guide the student through an application
help draft application responses using only information provided by the student
prepare application materials and email drafts for review

The AI acts as a copilot rather than making irreversible decisions. Important information such as GPA calculations, eligibility data, deadlines, and scholarship requirements is grounded in deterministic code or official university sources.

How we built it

Academic Copilot was developed on top of an existing Python scholarship-calculation project and expanded into a full AI-agent workflow during the hackathon.

We used Python and FastAPI for the backend, Selenium for retrieving academic information from the university portal, and HTML, CSS, and JavaScript for the interface.

DeepSeek powers the conversational agent. Rather than allowing the model to freely browse or calculate important information itself, we built a controlled tool system around it. The agent can call specific tools for retrieving the student profile, searching scholarships, ranking matches, projecting GPA, inspecting applications, and preparing application materials.

Scholarship information is gathered from official UPEI sources, while deterministic Python logic handles calculations and structured eligibility checks.

We also used OpenAI Codex extensively throughout development to accelerate implementation, debugging, testing, refactoring, and integration while we directed the architecture and product decisions.

Challenges we ran into

One of the largest challenges was dealing with university information that was designed for humans rather than software. Scholarship deadlines and application requirements can appear on different pages, inside application forms, or underneath broader award-cycle headings. Building a scraper that preserved this context was significantly harder than simply extracting text from individual scholarship pages.

Another challenge was deciding which responsibilities should belong to the AI model and which should remain deterministic. Allowing the model to calculate GPA, invent missing eligibility information, or guess deadlines would make the system unreliable. We therefore built clear boundaries between deterministic academic calculations, official-source retrieval, and the conversational AI layer.

We also had to make the system resilient enough for a live demo. External websites, browser automation, API limits, and login flows can all fail, so we maintained a demo mode and designed the application to fail gracefully.

Accomplishments that we're proud of

We are particularly proud that Academic Copilot became more than a chatbot placed beside a scholarship database.

The agent uses a real academic profile as context, independently selects tools, searches official sources, evaluates scholarship matches, asks for information it is missing, and continues through a multi-step application workflow.

We also built guardrails around the most sensitive parts of the workflow. Academic calculations remain deterministic, scholarship information retains its official source, the AI is not allowed to invent personal experiences for application essays, credentials are kept away from the AI model, and the student remains responsible for final submission.

The result is an agent that can move from "What scholarships should I apply for?" to a personalized shortlist and then directly into helping the student complete an application.

What we learned

The biggest lesson was that building a useful AI agent is less about giving a language model more freedom and more about giving it the right tools and boundaries.

We learned that deterministic software and generative AI work best together. Traditional code is better for calculations, validation, authentication, and structured eligibility rules, while the AI model is better at interpreting user requests, deciding which tools to use, explaining results, asking follow-up questions, and assisting with written applications.

We also learned how important source transparency is. For something such as scholarships, an AI answer is much more useful when the student can immediately see where the deadline or eligibility requirement came from.

What's next for Academic Copilot

The next major step is to turn Academic Copilot into a modular platform that can support any university rather than being tied to UPEI.

Instead of rebuilding the application for each institution, universities could have their own modules defining how academic records, scholarship directories, application systems, grading scales, and award requirements are accessed.

The long-term goal is for a student to connect their university and have Academic Copilot automatically adapt to that institution while keeping the same scholarship discovery, eligibility analysis, academic planning, and application-assistance experience.