bldr

BLDR: Repository for HackKU 2025

Update: We are on track to deliver this produect to University of Kansas' Advisor, and our tool will be used by students for schedule design from Fall 2025.

Inspiration

Our university’s only class scheduling tool was taken down right before fall and summer enrollment opened, leaving us to design our semester class schedule on our own, trying out different combinations of class time to fit our classes into schedule. This left us making our schedules on spreadsheets and papers. We knew there had to be a better way and we came across the idea of bldr.

What It Does

bldr is an AI-driven scheduling assistant that:

• Parses your academic advising report or unofficial transcript
• Suggests an optimal set of courses for your degree path
• Lets you refine or rebuild schedules via natural-language chat
• Syncs your final schedule to Google Calendar and email

Key Features

• Transcript Parsing
  Automatically detect which courses you’ve completed from PDF or image uploads.
• Gemini-Powered Chatbot
  Talk to bldr as if it were your academic advisor—impose constraints, ask for suggestions, or build a full schedule in one sentence.
• Google Calendar Integration
  One-click sync exports your schedule to your calendar of choice.
• Live Class Data Scraping
  Real-time section availability pulled from your school’s portal with Selenium & BeautifulSoup.
• Multi-Schedule Management
  Build, compare, and switch between multiple semester plans.
• Schedule Sharing
  Instantly email schedules to yourself or friends.
• Constraint-Based Planning
  Avoid 9 AM classes, limit credit hours, or block specific days with a simple prompt.

How We Built It

Frontend

• Next.js for SSR/SSG and routing
• Tailwind CSS & Shadcn UI for rapid, responsive styling
• Clean, mobile-first design powered by Framer Motion animations

Backend

• Express.js REST API
• PostgreSQL for structured data (course catalog, user profiles)
• MongoDB Atlas for flexible, document-based schedule storage

Parsing & Automation

• Selenium & BeautifulSoup scrape live course listings
• Custom parsers normalize disparate HTML structures into clean JSON using input of your class and semester selection

LLM Integration

• Google Gemini API powers our chatbot:
  • Ingests user context: completed courses, major requirements, current schedule
  • Differentiates passive vs. active intents (e.g., “What should I take?” vs. “Build my schedule”)
  • Calls our own endpoints (suggestedClasses(), addClassToSchedule(), etc.) to modify schedules agentically

Calendar & Email APIs

• Google Calendar API for one-click schedule export
• SendGrid (or similar) for instant email sharing

Challenges We Ran Into

• Web Scraping
  Inconsistent HTML required robust selectors and graceful failure handling.
• Conflict Detection
  Building a rules engine that respects time-block constraints, credit limits, and user preferences was non-trivial.
• Chatbot Contextuality
  Teaching Gemini to carry state—so it knows what you’ve already taken and your current plan—required careful prompt engineering.
• Agentic Features
  Enabling Gemini to actually call our APIs and update MongoDB documents on the fly was a complex orchestration.

Deep Dives

Using MongoDB Atlas

• Dynamic Schedule Storage
  Each schedule is a document with fields like schedID, semester, scheduleName, and a nested array of course objects.
• Schema Flexibility
  Perfect for handling multiple sections, temporary swaps, and AI-generated suggestions without rigid joins.
• Seamless Updates
  Fine-grained modifications (add/remove/replace classes) with single-field updates.

Using Google Gemini API

• Context-Aware Prompts
  Sends completed courses, major/catalog year, current schedule, and available classes in each request.
• Intent Classification
  Distinguishes between passive queries (“What do you recommend?”) and active commands (“Add ECON 101 at 10 AM”).
• Agentic Behavior
  On “active” intents, Gemini triggers our backend endpoints automatically, then confirms changes to the user.