CareFree

CareFree is a full-stack web application for student psychological support. It combines an AI-assisted first-contact flow, verified psychologist profiles, appointment requests, calendars, journaling and post-session feedback into one application.

The project is built as a monorepo with a Django REST backend and a Next.js frontend. It was developed as a practical health-tech prototype with an emphasis on privacy, clear user flows and responsible use of LLMs in a sensitive domain. CareFree is not a replacement for therapy, diagnosis or emergency support; the AI assistant is designed as an initial support and triage layer that helps users structure their situation and move toward human help.

What It Does

For students:

• register and complete a student profile
• talk with Julija, an AI assistant for initial support and triage
• receive psychologist recommendations based on the conversation context
• search approved psychologists by name and help categories
• request an appointment from a psychologist profile
• view confirmed appointments and Google Meet links in a calendar
• keep a private encrypted journal
• submit private post-session feedback

For psychologists:

• complete a professional profile with photo, CV, diplomas and certificates
• choose help categories and work approach
• define available one-hour appointment slots
• receive appointment requests from students
• approve or reject requests
• view confirmed appointments in a calendar

For administrators:

• verify psychologist profiles
• approve or deny psychologists
• manage users, categories and stored records through Django admin

AI And NLP Layer

The main AI component is Julija, a Croatian-language assistant implemented through the OpenAI API. Julija is not a generic chatbot; the backend treats the model response as structured application state.

The assistant flow includes:

• prompt-engineered conversation modes: support, recommendation offer, recommendation ready, support closure and crisis
• strict JSON responses parsed into a Pydantic schema
• category and subcategory extraction mapped to internal help-category codes
• session summaries that can be reused as context in later conversations
• psychologist recommendations based on approved profile categories
• fallback heuristics when the model call fails, times out or lacks enough context
• crisis-oriented behavior with local emergency contacts and separate fallback logic
• redaction of obvious sensitive identifiers before sending user text to the model

The journal module also includes a safety layer. Journal entries are encrypted at rest, and the application can classify potentially high-risk entries through a combination of rule-based checks and an LLM classifier.

Tech Stack

Backend:

• Python 3.12
• Django 5
• Django REST Framework
• SimpleJWT and cookie authentication
• Celery-ready background task structure
• OpenAI Python SDK
• Google Calendar API
• Fernet encryption through cryptography
• optional PostgreSQL, Resend email and Backblaze B2 media storage

Frontend:

• Next.js 16
• React 19
• TypeScript
• SWR
• Tailwind CSS 4
• React Big Calendar
• Radix UI primitives and lucide icons

Repository Structure

.
├── backend/                 Django REST API and domain modules
│   ├── accounts/            authentication, onboarding, roles, uploads
│   ├── users/               profile endpoints, search and help categories
│   ├── assistant/           Julija sessions, prompts, LLM integration, recommendations
│   ├── appointments/        requests, appointments, availability, feedback
│   ├── calendar_integration/ Google Calendar and Meet integration
│   └── journal/             encrypted journal and safety checks
├── frontend/                Next.js application
│   ├── app/                 routes and pages
│   ├── components/          shared UI and feature components
│   ├── fetchers/            API clients and tests
│   └── lib/                 auth, config, calendar and cache helpers
├── demo_profiles/           demo psychologist images used by seed scripts
├── scripts/                 local helper scripts
├── docs/                    architecture, setup, testing, research and operations docs
└── Dockerfile               backend deployment image

Internal backend model names still use caretaker in several places. The product-facing UI and documentation use psychologist or psiholog where appropriate.

Local Setup

Backend

cd backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
python manage.py migrate
python manage.py seed_help_categories
python manage.py runserver

For local development without external services, the minimal backend environment is:

APP_ENV=development
SECRET_KEY=change-me-for-local-dev
DEBUG=True
FRONTEND_URL=http://localhost:3001

If DATABASE_URL is not set, the backend uses local SQLite. For the full feature set, configure OpenAI, email, Google Calendar and optional B2 media storage as described in External Services.

Frontend

cd frontend
pnpm install
cp .env.example .env.local
pnpm dev

The frontend development server runs on:

http://localhost:3001

The backend development server defaults to:

http://localhost:8000

Demo Data

The repository includes demo profile images and seed tooling for local demos.

DEMO_CARETAKER_PASSWORD=<demo-psychologist-password> \
DEMO_STUDENT_PASSWORD=<demo-student-password> \
bash scripts/seed_demo_caretakers.sh

The seed command creates approved psychologist profiles, help-category assignments, availability, demo students, appointment requests and sample feedback. Generated credential snapshots are intentionally ignored by Git and should be shared only out of band.

For a full local reset, see Local Setup And Seeding.

Tests

Backend:

cd backend
./.venv/bin/python manage.py test

Frontend:

cd frontend
pnpm test -- --runInBand

Current test scope and caveats are documented in Testing.

Documentation

Start with docs/README.md for the full documentation map.

Key documents:

• Architecture
• AI Assistant
• Deployment
• External Services
• Testing
• Research Materials
• Ethics Materials

Security Notes

Do not commit .env files, API keys, OAuth secrets, local database files, generated credential snapshots or uploaded media. Production deployments must use a strong SECRET_KEY, a stable ENCRYPTION_KEY, HTTPS-only cookies, explicit CORS/CSRF origins and provider secrets stored in the deployment platform.

Journal content is encrypted at rest, but operational access, backups and key management still matter. LLM redaction is a minimization layer, not a full anonymization guarantee.

Project Status

The main product flows are implemented and documented: authentication, onboarding, psychologist verification, AI assistant sessions, recommendations, search, booking requests, Google Calendar/Meet integration, calendars, availability, journal and feedback. The project is suitable as a developed prototype, demo system and research artefact, with remaining work mainly around final QA, production hardening, broader clinical validation and long-term operational policies.