BurnBright | Devpost

Inspiration

As students juggling classes, projects, and responsibilities, we often found ourselves on the edge of burnout without even realizing it. Traditional planners don’t account for mental load, and students tend to normalize stress. We wanted to build a solution that not only tracks your academic workload but also proactively tells you when you’re likely to burn out and how to prevent it. That’s how BurnBright was born.

What it does

BurnBright is an AI-powered burnout prediction dashboard for students. It connects to your Canvas account, analyzes your upcoming assignments, quizzes, and deadlines, and

Predicts your burnout risk (0–100%) Identifies key stress triggers Suggests time and stress management strategies Recommends daily wellness habits Highlights your most stressful day of the week

How we built it

Frontend: Built using Next.js, TypeScript, TailwindCSS, and UI for styling.
Authentication: Used Auth0 for secure Google login.
Backend: A Flask (Python) API handles prompt formatting and workload analysis.
LLM Integration: We used Cerebras LLaMA-4 Scout to process structured prompts and generate multi-section wellness advice.
Database: MongoDB Atlas stores assignment data per user.
Canvas API: Retrieves real-time academic data from students' Canvas accounts.
We crafted structured prompts with tabular Canvas data and used custom headers to allow predictable, sectioned LLM responses.

Challenges we ran into

Canvas API's OAuth2 flow was difficult to automate in the time frame, so we temporarily used access tokens. Getting the LLM to return reliably structured output required prompt experimentation and careful formatting. UI layout needed to strike a balance between simplicity and informative insights. Managing asynchronous data fetching and syncing with LLM response timing in the frontend

Accomplishments that we're proud of

Delivering AI burnout predictions with zero model training, powered purely by smart prompt engineering. Successfully parsing real Canvas data and generating personalized feedback. Designing a smooth and helpful user experience tailored for stressed students. Separating LLM output into five clear sections for dynamic frontend rendering.

What we learned

How to use large language models beyond chat: for structured reasoning, analysis, and feedback. Importance of prompt engineering and output parsing for reliability. Frontend design matters just as much as backend logic when building for users under stress.

What's next for BurnBright

Integrating Google Calendar for combined academic and personal scheduling awareness. Visualizing burnout trends over time via graphs and heatmaps. Creating smart notifications and wellness nudges when burnout risk spikes. Working with campus wellness offices to deploy BurnBright as a student support tool.