Clarity AI

Introduction

Burnout is becoming increasingly prevalent, especially among college students. Constant tasks, overlapping commitments, and the pressure to excel create sustained stress that often goes unnoticed until productivity, motivation, and well-being decline. According to Handshake, 60% of college students experience some level of academic burnout. Burnout is strongly linked to reduced sleep; data from the National College Health Assessment shows that over 75% of students get fewer than eight hours of sleep on weeknights, a pattern that exacerbates stress and fatigue.

Common causes of burnout among students and professionals include:

Overload: Balancing classes, extracurriculars, jobs, and social commitments with little recovery time.
Perfectionism and pressure: Fear of failure, imposter syndrome, and unrealistic expectations drive sustained stress.
Neglected well-being: Poor sleep, skipped meals, lack of physical rest, and minimal self-care compound mental fatigue.

Our Inspiration

Burnout is difficult to detect because it develops gradually rather than appearing all at once. Early signs, such as fatigue, procrastination, or reduced enthusiasm, often blend into normal daily stress, making them easy to dismiss. Many students continually overload their schedules, unaware that their productivity and focus are slowly declining. Despite rising awareness of mental health challenges, digital solutions have remained reactive, focused on recovery after burnout rather than prevention.

For this reason, we decided to create an AI-agent based web app, Clarity AI, that can help people detect burnout in its early stages. The application integrates with Google Calendar, detecting common signs of burnout like tightly packed schedules, long stretches without breaks, and an abundance of high-stress activities.

Key Features

Securely connects to the user’s Google Calendar through OAuth, allowing it to automatically analyze upcoming and past events without requiring manual data entry.
Calculates a personalized stress score based on the user’s schedule over a rolling 7-day period.
Evaluates event titles and descriptions to estimate the stress level of each activity.
Measures how packed a user’s week is by examining the number of events per day, the average length of breaks, and how frequently events overlap, flagging unsustainable scheduling patterns.
Analyzes the largest available sleep window based on the user’s schedule.
Uses Claude Sonnet 4.5 to identify habits that may lead to burnout and provides personalized recommendations to adjust their calendar, helping them reduce stress and balance their workload.

Stress Assessment Metrics

Calendar Density

Percentage of waking hours occupied by scheduled events during a 7-day period
(Total scheduled minutes / Waking minutes) * 100

Interpretation

0-20%: Light schedule, ample free time
21-40%: Moderate schedule, balanced workload
41-60%: Heavy schedule, limited flexibility
61-80%: Very heavy schedule, minimal free time
81-100%: Severely packed schedule, burnout risk

Sleep Hours Available

Average nightly sleep opportunity based on schedule gaps during typical sleep windows
(Total sleep hours / 7 days a week)

Interpretation

8+ hours: Excellent sleep opportunity
7-7.9 hours: Good, within recommended range
6-6.9 hours: Insufficient, below minimum
4-5.9 hours: Severely deprived, major cognitive impact
<4 hours: Critical, seek immediate support

Immediate Action Items

Number of tasks due today or tomorrow

Interpretation

0-1 tasks: Low immediate pressure
2-3 tasks: Moderate deadline stress
4-5 tasks: High immediate pressure
6+ tasks: Critical deadline overload

Events Over Next Week

Total number of scheduled events in the upcoming week

Stress Level of Events

Event type significantly impacts perceived stress (exam vs. social activity). The stress level is determined by analyzing key words in the event name and description.

High-stress events: +4 points each
Recreational events: -2 points each (stress relief)
Neutral events: +1.2 points each (base weighting)

Architecture Overview and AWS Tools

Frontend and Backend

The web app is built using Next.js, which provides a responsive and dynamic interface for users to view their calendar insights and AI recommendations.
The backend is implemented in Python, handling calendar data retrieval, stress score calculations, and communication with AWS Bedrock for AI queries.

Generative AI Use

We used Claude Sonnet 4.5 via AWS Bedrock for AI insights
Two types of AI insights: predictions and interventions
First, the AI assistant was asked to predict potential high-stress moments and burnout periods for the user.
- These were based on “stress factors” including the number of events in the following week, the calendar density, the sleep hours available, and immediate action items
- The assistant was required to give predictions in a structured JSON format
The AI assistant was separately asked to generate interventions and feedback for the user
- This was based on the stress score and the events listed

Benefits of Using AWS

Streamlined process, multiple models found in one place
- Allowed for easy testing of multiple models on their performance
Security/encryption - A significant advantage of AWS is that all data stored within AWS Bedrock is automatically encrypted at rest using AWS Key Management Service. All network traffic to and from AWS Bedrock endpoints uses TLS to ensure encryption in transit, so all prompts, model responses, and metadata are secure.

Challenges Faced and What We Learned

Defining burnout: Translating qualitative well-being concepts into measurable patterns was challenging. We considered key factors from Google Calendar, like occupied time and estimated sleeping hours, to quantify stress.
AWS tools: Before the summit, none of us were well-versed in AWS’s GenAI offerings. Through the workshops, we learned how to Bedrock to efficiently build powerful tools leveraging LLM capabilities.
Personalization: The AI Agent needed to adapt predictions and recommendations to each user’s schedule. We crafted detailed prompts requiring it to reference specific events and tasks, ensuring actionable, tailored insights.

Impact and Future Vision

Clarity AI empowers users to take care of their wellness without extra effort–self-awareness built into their workflow. It also empowers counselors to care for their students and truly know how a student may be doing, whether the student lies or not.
Next Steps:

Integrate more apps: Considering data like Notion, Gmail, and even GitHub may provide further
Add long-term trend analysis: Look at the change in patterns over months for sustained behavioral tracking
Expand dashboard: Expanding available information to audiences such as counselors, as well as predictive analytics