Inspiration
Mental wellness is an important part of student life, but students may hesitate to seek help or may not know where to start. We wanted to build a system that could provide an accessible first step while still keeping professional human support at the center of the process.
This led us to build MindCare AI, a digital mental wellness platform designed specifically for students in higher education. The idea is to connect assessment, early support, human follow-up, counselling, and progress monitoring into one controlled workflow instead of treating them as separate activities.
What It Does
MindCare AI allows students to provide informed consent and complete structured wellness assessments. The system calculates a deterministic score and classifies the result into Low, Moderate, or High risk.
The documented workflow is:
- 0–9 → Low
- 10–19 → Moderate
- 20–30 → High
Low-risk interactions can receive bounded AI-assisted wellness support. Moderate-risk cases move through a structured Check-up Member and Tester workflow, while high-risk cases are routed toward professional counselling.
The platform also supports case management, progress tracking, reassessment, counselling sessions, support plans, notifications, and audit records.
How We Built It
We designed the platform using a microservices architecture.
The frontend uses React, Vite, and TypeScript. Requests pass through Spring Cloud Gateway and are routed to dedicated Spring Boot microservices.
The backend is based on Java 21 and Spring Boot. Separate services are planned for authentication, users, institutions, assessments, AI support, cases, check-ups, tester workflows, counselling, administration, notifications, and auditing.
Oracle Database 21c XE is used for persistent data. Redis is used where caching or temporary high-speed access is required, while Apache Kafka supports asynchronous domain events. Eureka provides service discovery, and Spring Cloud Config provides centralized configuration.
For AI functionality, we use Spring AI with a Large Language Model. The AI is intentionally designed as an assistive component rather than an autonomous diagnostic system.
AI and Safety
A major part of the project was designing AI responsibly.
The AI receives only the permitted information required for the support interaction. Its output is processed through safety and privacy controls before being shown to the student.
The AI cannot change deterministic assessment scores or risk categories. It also cannot independently close cases or make clinical decisions.
When the AI service is unavailable, times out, or produces unsafe or malformed output, the system is designed to provide a safe fallback and direct the user toward appropriate human support.
This human-in-the-loop approach keeps professional staff and counsellors responsible for important support and escalation decisions.
Challenges
One of our main challenges was designing the workflow for different risk levels while keeping the system safe and understandable.
Another challenge was separating responsibilities across multiple microservices while maintaining communication between them. Authentication, assessment, case management, counselling, notifications, and auditing each have different responsibilities and access boundaries.
AI integration also introduced additional challenges because generated responses cannot simply be trusted as medical advice. We therefore designed explicit safety rules, fallback behavior, privacy controls, and human escalation.
Security and privacy were also major considerations because the platform handles sensitive student wellness information. The design therefore includes authentication, role-based authorization, consent controls, HTTPS, secret management, least privilege, and auditability.
What We Learned
Through this project, we learned how to design a full-stack application around real-world workflows rather than only individual features.
We gained experience with Java, Spring Boot, Spring Cloud microservices, REST APIs, React, TypeScript, Oracle Database, Redis, Kafka, Docker, and Spring AI.
We also learned that adding AI to a sensitive application requires more than connecting an LLM to a chatbot. The surrounding system needs clear boundaries, privacy protection, fallback mechanisms, deterministic rules where appropriate, and human oversight.
Future Improvements
Future development can include stronger implementation-level AI evaluation, improved monitoring, additional accessibility features, more comprehensive testing, and further integration with institutional support workflows.
The system can also be expanded while preserving its core principle: AI should assist student wellness workflows while qualified human professionals remain responsible for appropriate escalation and counselling.