Classly

Inspiration

As college students taking technical courses, we constantly found ourselves switching between Canvas, Gradescope, PrairieLearn, Campuswire, and scattered PDF syllabi. Deadlines lived in different systems, grading breakdowns were buried in documents, and study guidance required manual digging. We wanted a single academic control center that didn’t just display information—but acted on it.

What It Does

Classly is a unified coursework dashboard that centralizes assignments, grades, announcements, and schedules across platforms.

It also includes a course-specific AI assistant powered by RAG. Students select a class and ask natural language questions like:

“What upcoming assignments do I have?”

“How should I study for my quiz?”

“What’s the grading breakdown?”

Beyond answering questions, Classly includes an AI agent that automatically creates Google Calendar events when deadlines are detected. It schedules optimized study blocks and reminders based on your availability.

How We Built It

We scraped and structured course data from Canvas, PrairieLearn, Campuswire, and official course sites using Selenium. Parsed materials (syllabi, assignments, announcements) are normalized and stored in Supabase Postgres.

We use pgvector to store embeddings for semantic search. Our RAG pipeline generates query embeddings at runtime, retrieves the most relevant chunks scoped strictly to the selected course, and feeds grounded context into the LLM to produce accurate responses.

Using the Keywords AI LangChain SDK, we built agentic workflows that detect scheduling intent, extract deadlines, check Google Calendar availability, and automatically create optimized events. Prompt versioning through Keywords AI ensures consistent behavior across retrieval and planning agents.

Challenges We Ran Into

One major challenge was grounding responses strictly within the selected course to avoid cross-course contamination or hallucinations.

We also had to carefully design the agent workflow to prevent duplicate calendar events and ensure scheduling logic respected real availability constraints.

Finally, normalizing scraped data across multiple LMS platforms required building a consistent schema before embeddings could be reliably generated.

Accomplishments That We’re Proud Of

We successfully built an end-to-end academic assistant that not only answers course-specific questions but also takes real-world actions.

We integrated scraping, vector search, RAG, agent workflows, and Google Calendar automation into one cohesive system.

Most importantly, Classly transforms static class content into a proactive scheduling system rather than just another dashboard.

What We Learned

We learned that RAG alone is not enough—agent orchestration and clear scoping are critical for reliability.

We also saw how powerful prompt management and workflow control become when building multi-step AI systems.

Most importantly, we learned how to move from “AI that chats” to “AI that acts.”

What’s Next for Classly

Next, we plan to:

Add real-time LMS API integrations instead of scraping

Introduce cross-course workload optimization

Add adaptive study planning based on performance trends

Expand to multiple universities with scalable onboarding

Our vision is to make Classly the default AI academic operating system for students everywhere.

Built With

• javascript
• keywordsai
• langchain
• lovable
• postgresql
• python
• selinium
• supabase
• trae
• typescript