Mental Health Helper

We aim to help people with overwhelming pressure.

Comment
  1. What Inspired Me

My inspiration for this project came from researching Cornell’s student mental-health situation, especially while trying to understand suicides on campus over the past five years. During this research, I found that:

Cornell does not publish clear annual suicide numbers, largely due to privacy and family wishes.

The university only releases general trends, such as “student suicide rates have declined since 2010,” without specific figures.

Mental-health surveys, however, showed troubling data—

around 9% of Cornell students seriously considered suicide,

roughly 85 students per year reported a suicide attempt, and

over 40% experienced depression or anxiety severe enough to impair functioning.

Seeing this gap—high psychological distress vs. lack of precise, transparent data—made me realize how urgently students need accessible, real-time emotional support. This directly inspired me to build an AI-powered CBT therapy agent that could serve as an always-available, compassionate companion for students struggling with overwhelming emotions.

Ultimately, the motivation came from a simple question:

If accurate suicide numbers are so difficult to obtain, how many students suffer silently long before a crisis ever appears in the statistics?

I wanted to build something that could help before someone reaches that point.

  1. What I Learned

Through this project, I gained insights on several levels:

Psychological & Clinical Insight

Suicide statistics are often incomplete, but self-harm ideation rates show the true scale of psychological distress.

CBT (Cognitive Behavioral Therapy) follows a structured framework that can be modeled into a step-by-step conversational agent.

Real mental-health support requires not just conversation, but emotion detection, risk assessment, and pattern tracking.

Human Factors & Ethical Understanding

Mental-health systems must be built with privacy, sensitivity, and safety as core priorities.

Even advanced AI cannot replace therapists, but it can extend their reach and help detect early warning signs.

Technical Learning

Integrating multiple components—LLMs, speech-to-text, data pipelines, risk detection—requires careful architectural planning.

Real-time dashboards, auto-refreshing user views, and trend visualization dramatically improve therapist usability.

Model outputs must be validated to avoid harmful responses.

  1. How I Built the Project

We developed the system with two distinct interfaces:

(1) Patient-facing AI Therapy Agent

“Calm-Tech” UI design to minimize cognitive load

Google Speech-to-Text API for voice input

Structured CBT flow:

presenting problem → emotions → physical reactions → automatic thoughts → behaviors → consequences → reframing

Real-time analysis immediately after each message

Session history stored securely for long-term progress tracking

(2) Therapist Dashboard

Real-time patient monitoring (refresh every 3 seconds)

Automatic clinical scoring:

depression, anxiety, stress, rumination, avoidance, self-blame

Cognitive distortion detection

Emotional + behavioral pattern mapping

Radar charts, bar graphs, and trend visualizations

Patient profile pages with longitudinal analytics

Tech Stack

LLMs for analysis + therapeutic dialogue

Next.js / React for frontend

FastAPI / Node.js backend

Google Speech-to-Text

SQL / Supabase for storage

Real-time update engine via WebSockets

The entire system was built with one goal: make psychological support accessible, data-driven, and immediate.

  1. Challenges I Faced

This project presented several major challenges:

(1). Handling Sensitive Topics Responsibly

While researching Cornell suicide patterns, I realized how easily data can be misinterpreted. The biggest challenge was designing the system so that:

It never makes unsafe statements

It can triage risk levels accurately

It can respond appropriately to self-harm language

(2). Designing a Safe but Supportive AI

Creating the right tone—empathetic but not overstepping clinical boundaries—required multiple prompt iterations, guardrails, and fallback safety layers.

(3). Real-time Data Pipeline

Ensuring the therapist dashboard auto-refreshes every 3 seconds while maintaining privacy and performance took careful engineering.

(4). Balancing Clinical Structure with Human Conversation

CBT is structured, but humans are not. Building a flow that feels natural yet clinically valid was surprisingly difficult.

(5). Turning Research into a Working System

The Cornell suicide data challenge taught me something important: real-world mental health problems rarely come with clean data or clear definitions. Our system had to be flexible enough to work with uncertainty.

  1. Final Reflection

This project was driven by a simple belief: students deserve timely, compassionate, and intelligent mental-health support—especially when real-world data hides the full extent of suffering.

By combining CBT structure, real-time analytics, and AI-powered conversation, we created a system that could genuinely help users feel seen, supported, and understood.

And personally, this project taught me that meaningful tech doesn’t start from code— It starts from empathy.

Built With

Share this project:

Updates