Pace

AI Burnout Prevention Platform

Comment

Inspiration

We've all been there: three exams in one week, back-to-back meetings with no breathing room, and the creeping realization that you haven't had a real break in days. Burnout doesn't announce itself—it accumulates quietly until you're too exhausted to function. We built Pace because existing productivity tools focus on doing more, not on sustaining yourself while doing it. We wanted a system that listens to how you're actually feeling, looks at your real schedule, and intervenes before you crash—not after.

What it does

Pace is a web-based burnout prevention platform that combines voice reflection, stress tracking, and calendar analysis to automatically design recovery time into your schedule.

Core workflow:

  1. Weekly Vent - Record voice reflections about how your week feels. ElevenLabs transcribes instantly, and AI provides immediate contextual insights.
  2. Stress Check - Interactive quiz tracks stress, overwhelm, and energy levels over time. Scores feed into AI decision-making.
  3. Pacing Advice - AI analyzes your voice memos, calendar density, and stress patterns to generate personalized pacing strategies (focus sprints, recovery rituals, connection blocks).
  4. Smart Calendar - The AI determines optimal break frequency and timing based on your workload and stress level, then automatically schedules recovery breaks directly into your Google Calendar—no manual input required.

Pace doesn't just suggest breaks. It schedules them for you, blocking time before burnout hits.

How we built it

Tech Stack:

  • Backend: Flask (Python) with Firebase Admin SDK for user authentication and Firestore database storage
  • Calendar Integration: Google Calendar API with OAuth 2.0 for syncing events and automated scheduling
  • Voice Processing: ElevenLabs Speech-to-Text API for instant transcription of voice reflections
  • AI Engine: Google Gemini powers three core LLM pipelines:
    1. Assistant Insights LLM - Generates contextual tips for vent and stress check pages
    2. Pacing Advice LLM - Analyzes voice memos, calendar density, and stress scores to create strategic workload plans
    3. Smart Break Scheduler LLM - Determines break frequency, timing, and type, then triggers automated Google Calendar event creation

Architecture:

  • User authentication via Google OAuth
  • Real-time audio recording and Firebase storage
  • Event data synced from Google Calendar and cached in Firestore
  • AI prompts engineered to reference specific calendar events, transcription sentiment, and stress trends
  • Automated scheduling using Google Calendar API's event insertion endpoint

Challenges we ran into

  1. Prompt engineering for personalization - Getting Gemini to reference specific calendar events and voice memo content (not generic advice) required iterative refinement of context injection and output formatting.

  2. Calendar time zone complexity - Handling Google Calendar's mix of dateTime and date formats, plus timezone conversions, while enforcing the 10pm-7am blackout window for break suggestions.

  3. Balancing automation with agency - Deciding when the AI should suggest vs. schedule breaks. We landed on auto-scheduling with user approval to maintain trust while reducing friction.

  4. Transcription quality - Voice notes are messy—pauses, filler words, emotional tone. Training the insights LLM to extract meaningful patterns from unstructured reflection data took experimentation.

  5. Firebase data modeling - Structuring user data (transcriptions, stress scores, calendar events) for fast queries while supporting historical trend analysis and real-time updates.

Accomplishments that we're proud of

  • It actually works end-to-end - Voice recording → transcription → AI analysis → automated calendar scheduling. The full loop runs without manual intervention.
  • Personalization that feels real - The AI references your actual events ("Your CS Project Presentation on Thursday...") and your own words from voice memos, making advice feel human, not templated.
  • Proactive intervention - Pace schedules breaks before heavy workload days, not after. The system anticipates burnout instead of reacting to it.
  • Clean integration - Google Calendar sync is seamless. Users don't need to learn a new calendar system—Pace enhances what they already use.
  • Privacy-first design - Voice transcriptions and stress data stay in user-scoped Firebase collections. No data sharing, no third-party analytics.

What we learned

  • AI needs context to be useful - Generic advice is noise. We learned to feed LLMs rich, specific context (calendar events, stress history, voice sentiment) to generate actionable recommendations.
  • Voice is powerful for reflection - Text input has friction. Voice notes capture emotional nuance and are faster to create, leading to richer data for AI analysis.
  • Automation requires trust guardrails - Users need to see why the AI suggested a specific break time. Transparency in reasoning builds confidence in automated scheduling.
  • Burnout is predictable - Patterns emerge: deadline clusters, back-to-back meetings, declining energy scores. AI can spot these patterns before humans consciously recognize them.
  • Flask + Firebase + Google APIs play well together - The combination gave us rapid prototyping speed with production-ready authentication and data storage.

What's next for Pace

  1. Mobile app - Native iOS/Android apps for faster voice recording and push notifications when high-stress periods are detected. We used Flutter, and have a finished UI for flutter, however the wifi networks have too many users on it(because of the hackathon), so running locally wasn't working. We do plan to deploy the backend to avoid this for the app launch.

  2. Multi-calendar support - Handle work, personal, and academic calendars separately to account for different energy costs per domain.

Built With

Share this project:

Updates