The inspiration for this project came from a simple observation: students are overwhelmed.
They juggle:
Classes
Assignments
Exams
Projects
Weekly planning
Random reminders scattered across WhatsApp and screenshots
Despite this, most universities do not provide a simple, unified planner that students can ask questions like:
“What classes do I have today?” “Make me a weekly study plan.” “What subjects am I taking this trimester?”
I wanted to build an assistant that felt alive, helpful, and instant — something that gives structure to chaos. Even without real student data, I wanted the AI to generate realistic academic information, fully fictional but believable, and perfectly formatted for students.
The idea was inspired by:
The flexibility of modern LLMs
The creativity of prompt engineering
The goal of building a planner that feels like a personal academic secretary
🧠 What I Learned
This project taught me a surprising amount about LLM behavior, prompt design, and structured data simulation.
- The Power of Strict Prompt Structures
I learned quickly that LLMs do not act consistently unless you build a “behavioral rule system.” For example:
Clear categories for types of student questions
Priority rules (e.g., schedule → highest priority)
Allowed durations for classes
Forbidden behaviors (no tool calls, no real data)
This is similar to creating a deterministic rule engine using language.
- Building Realistic Fictional Data
Because the assistant does not use any real database, I learned how to generate believable academic content:
Subjects
Semesters
Assignments
Exams
Grades
With constraints such as:
Class Duration Constraint
Each class must be either 1 or 2 hours:
duration ∈ { 1 , 2 } duration∈{1,2} Non-overlapping Scheduling Rule
If a class ends at 𝑡 𝑖 , end t i,end
, the next class must start at or after that time:
𝑡 𝑖 + 1 , start ≥ 𝑡 𝑖 , end t i+1,start
≥t i,end
These math constraints helped me formalize how the planner should behave.
- Using Markdown and LaTeX for Clarity
I learned to structure responses using:
Markdown headings
Bullet lists
Tables
Code blocks
LaTeX math
This made the final project feel polished and professional.
🏗️ How I Built the Project
I built the project around a core system prompt that completely defines the assistant’s behavior.
Step 1 — Identify User Intent Categories
I mapped every possible student query into categories:
Daily schedule
Weekly timetable
Subjects
Semesters
Assignments & deadlines
Exams
Grades
Academic advice
This allowed the assistant to behave deterministically.
Log in or sign up for Devpost to join the conversation.