home page
logo

Inspiration

Many professors will assign homework in sites outside of canvas, much to the annoyance of students. What's more, professors might post their syllabus over multiple files for different sections of the class. When you're a college student, keeping track of all the different sites, dates, assignments, and resources gets hard quickly.

What it does

This is where Office Hours comes in. With a single prompt, get a detailed summary of your upcoming assignments, both from canvas and elsewhere. Have the Office Hour Assistant build you a time schedule to balance study and campus events, create a study sets, and search for where your professor hid that extra homework assignment.

How we built it

We used NVIDIA's Nemotron AI model to manage a swarm of Claude agents that would scrape canvas and the rest of the web to find the relevant information for your courses and your prompt and store it into a Tiger Data database. Nemotron then summarizes the data on your assignments and events, and sends it to another Claude agent to turn into embedded HTML. We used the timestamps from Tiger Data's hypertables to be able to check when the Claude agents needed to be told to scrape for new data, or if Nemotron was able to just use the already existing data.

Challenges we ran into

When it came to having a refined front-end talking to a swarm of AI agents on the backend, integration was easily one of our biggest challenges. Having Nemotron summarize its data and then actually getting Claude to turn that data into render-able HTML took us at least 2 hours. Our next biggest challenge came in the testing phase: Nemotron would often fail to find the sandbox canvas environment we gave it, and would refuse to crawl through links unless we explicitly whitelisted them beforehand.

Accomplishments that we're proud of

The workflow of managing multiple AI agents using a single leader is a sophisticated step towards modern agentic AI processes. Additionally, our front end graphic design is sleek and professional looking, and any live-generated elements consistently match the rest of our theme.

What we learned

We learned how to effectively deploy a swarm of agentic AI processes to scrape the web and summarize data. Additionally, we learned how to setup and use NVIDIA Nemotron and it's powerful, tool-based workflow.

What's next for Office Hours

Next steps for Office Hours include deploying to a live environment like Vercel so that end users can begin to test our application. We also need to create a proper system for calling the canvas API, and request that universities whitelist our application as an approved API consumer.