UniversityCalendarSync AI

Sync your university schedule with Google Calendar, effortlessly

Comment

Inspiration

Every semester, students face the same frustrating ritual: staring at a schedule posted on a wall, a screenshot from a chat, or a clunky university portal — and then manually typing each class into Google Calendar. We've all been there. UniCal was born from a simple question: why can't a computer just look at the schedule and do this for us?

What it does

UniCal takes a URL to any university schedule page and automatically:

  1. Navigates the site using an AI agent that acts like a real student browsing the portal
  2. Extracts the schedule by reading the page content with LLM-powered parsing
  3. Normalizes messy data (different languages, formats, abbreviations) into structured events
  4. Exports directly to Google Calendar with proper recurrence rules for the entire semester

The user just pastes a link, picks a semester start/end date, and clicks import. That's it.

How we built it

The system has three main components running in Docker:

  • Browser Service — a headless Chromium controlled by an AI navigator. It receives screenshots + DOM elements and decides where to click, scroll, or navigate, mimicking a real student looking for their timetable. Once found, a separate LLM-powered extractor parses raw HTML into structured JSON (day, time, subject, room, teacher, type).
  • Web App — an Express.js server with Google OAuth, Prisma ORM for PostgreSQL, and a static frontend. It orchestrates the extraction flow and handles Google Calendar API integration, creating recurring events with proper RRULE definitions.
  • Infrastructure — Docker Compose for local dev, Caddy for HTTPS in production, and a CI-friendly build/deploy pipeline.

We used OpenAI models (GPT-4o-mini for navigation, GPT for extraction) as the "eyes and brain" of the system. The extraction pipeline is a port of an earlier Python prototype into Node.js, preserving the chunking and deduplication logic.

Challenges we ran into

  • Navigating arbitrary university portals — every site has different structure, auth flows, and UI patterns. Getting the AI agent to reliably find the schedule page required careful prompt engineering and a robust action loop with screenshot feedback.
  • Schedule formats are chaotic — abbreviations, mixed languages (Russian/English), inconsistent time formats, merged cells. We built extensive normalization (day name mapping in both languages, flexible parsing).
  • Token budgets — some university pages have enormous DOM trees. We built HTML minimization and chunking logic to stay within LLM context limits while preserving schedule-relevant content.
  • Google Calendar recurrence — correctly mapping "every Tuesday at 10:00 for the semester" to RRULE with proper BYDAY, DTSTART, and COUNT took more debugging than expected.

Accomplishments that we're proud of

  • The AI navigator genuinely works on university portals it has never seen before — it reads the page, finds the schedule section, and gets there autonomously.
  • End-to-end flow in under 2 minutes: paste a URL → get a full semester of recurring events in your Google Calendar.
  • The HTML cleaning pipeline strips framework noise (React/Angular/Vue/Svelte attributes) so effectively that even massive pages fit within LLM context windows.
  • Bilingual support out of the box — Russian and English schedule formats are handled seamlessly.

What we learned

  • AI agents that interact with real websites need a tight observe → reason → act loop grounded in actual DOM elements, not just screenshots.
  • LLM-powered data extraction works surprisingly well on messy HTML when you strip framework noise and focus the model on semantic content.
  • The gap between "works on our test schedule" and "works on any schedule" is where most of the engineering effort lives.

What's next for UniversityCalendarSync AI

  • Support for image/photo input (OCR pipeline for schedules on whiteboards and paper)
  • Apple Calendar and Outlook export
  • Schedule change detection and automatic calendar updates
  • Group sharing — one student extracts, the whole group gets the calendar
  • Support for more languages and university systems worldwide

Built With

Share this project:

Updates