Boiler Buddy

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

Boiler Buddy Inspiration

College students face increasing mental health challenges due to academic pressure, social transitions, and limited access to timely support. At Purdue University, services like CAPS are essential but often in high demand, and many students hesitate to seek help due to stigma or lack of immediate access.

Boiler Buddy was created to provide an always-available, low-barrier support system that complements professional resources while prioritizing user privacy and safety.

What it does

Boiler Buddy is a real-time, AI-powered mental health support application designed specifically for university students. It combines conversational AI, voice interaction, behavioral analysis, and campus-specific resources into a single, privacy-first platform.

The system enables users to:

Engage in therapeutic-style conversations Speak naturally through voice input and receive spoken responses Reflect through guided journaling Access immediate crisis resources Monitor and manage stress through integrated wellness tools

All processing is done client-side where possible, ensuring that sensitive data remains private and under user control.

Key Features Core Chat Experience AI Therapeutic Chat powered by Anthropic Claude System prompt designed for person-centered support: Reflects emotions before giving suggestions Uses light CBT/DBT techniques Asks one open-ended question at a time Avoids diagnosis or medical claims Purdue-specific tone and encouragement Animated Avatar Interface Real-time animated 2D Purdue Pete SVG avatar 60 FPS animation using requestAnimationFrame Behavior includes: Mouth movement during speech Eye reactions while listening Idle breathing and blinking Arm animation with hammer Voice Interaction System Speech-to-text via Web Speech API Live transcription during user speech Automatic message submission after recording Text-to-speech output with dual backend: ElevenLabs premium voices (Adam, Josh, Rachel, Sam, Bella, Antoni) Browser voice fallback (Microsoft/Google voices) Voice selector with preview functionality Safety and Crisis Response Real-time crisis keyword detection Immediate interruption of conversation flow Full-screen crisis modal with direct actions: Purdue CAPS 24/7 Crisis Line Crisis Text Line (741741) 988 Lifeline Privacy and Security No Personal Health Information (PHI) stored externally All data stored locally in browser (localStorage) Stateless architecture with no backend database Clear Session and Clear All Data functionality Informed consent screen requiring acknowledgment of: Non-therapeutic nature of the app Data privacy limitations Emergency protocols Quick Exit (Dorm Privacy Feature) Panic keyword ("Boiler Up") or triple Escape triggers instant screen swap Displays a fake Purdue Brightspace interface Hidden return mechanism Voice Journaling Dedicated journaling mode using voice input Automatic transcription of spoken entries AI-generated reflective summaries Follow-up prompts for deeper reflection Entries stored locally with mood indicators Wellness Tools Pomodoro-based Study Break Timer: Study (25 min) Break (5 min) Breathing (1 min) Context-aware suggestions when stress is detected Resource Hub: Purdue CAPS contact and location 24/7 crisis services 988 Lifeline Crisis Text Line TAO Connect platform One-tap call and text actions Acoustic and Behavioral Intelligence Real-time voice stress analysis using Web Audio API Detects: Energy levels Monotone speech Voice tremors Pauses Displays proactive stress alerts with intervention suggestions Voice biometrics ("Roommate Shield"): Optional speaker verification FFT-based frequency comparison with cosine similarity Prevents unauthorized interaction Usage Monitoring Tracks: Sessions per week Total interactions Consecutive usage days Detects over-reliance and encourages seeking professional help Session Management AI-generated session summaries with mood indicators Local session history with timestamps and message counts Ability to clear all history at any time How we built it

Boiler Buddy is built using a modern, client-focused web architecture:

Frontend: Next.js 16 (App Router) with TypeScript Styling: Tailwind CSS v4 Animation: Framer Motion + requestAnimationFrame AI: Anthropic Claude (claude-sonnet-4) Text-to-Speech: ElevenLabs API + SpeechSynthesis fallback Speech-to-Text: Web Speech API Audio Processing: Web Audio API (FFT + AnalyserNode) Storage: localStorage only (no backend database) Challenges we ran into

One of the main challenges was maintaining a high level of personalization while enforcing strict privacy constraints. Designing a system that feels adaptive without storing sensitive data required a fully client-side architecture.

Another challenge was achieving natural voice interaction. While high-quality voice models were integrated, additional work was required to structure responses with natural pacing, emotional variation, and conversational flow.

Implementing real-time stress detection from audio also required careful tuning of signal processing techniques to avoid false positives while still providing meaningful feedback.

Accomplishments that we're proud of Built a fully functional, privacy-first mental health assistant with no backend data storage Designed a real-time animated avatar system with synchronized voice interaction Implemented acoustic stress detection using raw audio analysis Developed a multi-layered crisis detection and response system Integrated Purdue-specific resources into a seamless user experience What we learned

We learned that effective mental health tools require more than conversational AI. Features like journaling, structured reflection, and proactive interventions significantly improve usability and impact.

We also found that voice interaction quality depends heavily on response design, not just the underlying model. Small improvements in phrasing and pacing dramatically affect perceived realism.

Finally, we gained experience building privacy-first systems that handle sensitive use cases without relying on traditional backend infrastructure.

What's next for Boiler Buddy Improve voice realism and latency through streaming audio Expand behavioral intelligence with multi-modal inputs Add wearable integration for passive stress detection Develop optional HIPAA-compliant backend for secure long-term tracking Scale the platform for deployment across other universities Disclaimer

Boiler Buddy is not a licensed medical provider and does not provide diagnosis or treatment. It is intended as a supportive tool and resource guide. In emergency situations, users are directed to professional services.