AutiSmart
AI-assisted autism-support platform — team Final Year Project at FAST-NUCES (2025–2026).
1. The Problem
Parents and caregivers of children on the autism spectrum often need structured ways to track behaviours, complete stage-oriented assessments, and receive practical guidance between clinical visits. Existing tools are frequently fragmented: assessments live in one place, progress notes in another, and therapy activities elsewhere.
AutiSmart was designed as an academic software prototype that brings these workflows into a single full-stack application — assessment, child profiles, progress tracking, role-based access, therapy-oriented activities, and AI-assisted question generation / recommendations — so the team could demonstrate end-to-end product engineering in a healthcare-adjacent domain.
Scope boundary: This is a team Final Year Project and research-oriented prototype. It is not a clinically validated diagnostic system, medical device, or production healthcare platform.
2. The Solution
AutiSmart is a role-based web platform with the following verified capabilities (supported by repository models, services, routes, and documentation):
- Role-based authentication — caregivers, experts, and administrators (JWT + email OTP verification)
- Child profiles & caregiver management — structured child records linked to caregivers
- Structured assessment workflows — multi-category quizzes (Eye Contact, Social Interaction, Communication, Repetitive Behavior, Sensory Sensitivity, Focus & Attention) with result history and level classification
- AI-assisted quiz generation — Groq (Llama 3.3 70B) generates assessment questions and personalised child quizzes
- Therapy-game recommendations — rule-based ranking of activities from latest assessment category scores
- Emotion Explorer — AI-generated emotion scenarios and encouraging feedback for therapy-oriented play
- Progress tracking & reporting — statistics, history, and dashboard-oriented views
- Administrative controls — user management, assessment management, platform statistics
Features such as multimodal audio/video analysis or production clinical pipelines are not claimed; they are outside the verified implementation.
3. System Architecture
Classic monolithic full-stack architecture. No microservices or complex infrastructure were introduced.
Stack: React 19 + Vite 7 frontend · Express.js / Node.js backend · MongoDB + Mongoose · JWT + RBAC · Groq API · Nodemailer
Key data models: User (role, OTP), Child, Assessment, AssessmentResult, ChildQuiz, Activity.
External services actually integrated: Groq API (AI generation), Nodemailer (email OTP). Firebase/Twilio appear in dependencies but are not required for the core documented flows.
4. Team Project vs My Contribution
This is a team Final Year Project. The public repository and this case study represent collective work. Individual ownership of every line of code is not claimed.
My Documented Role
Full-Stack Development & AI Integration
- Full-stack work across the React frontend and Node.js/Express backend
- AI integration using the Groq API (quiz generation, Emotion Explorer scenarios & feedback)
- Authentication and role-based workflows (JWT, OTP via Nodemailer)
- Contribution to system architecture, REST APIs, and product workflows
Teammates (from project documentation)
- Shayan Ahmad — Core Engine & Backend Development
- Ahmad Kamran — Frontend UI & Systems Logic
Final presentation and academic grades (FYP-I A+, FYP-II A-) reflect the team outcome.
5. Key Technical Decisions
React + Vite frontend
Why: Fast development iteration, component model suited to dashboards and multi-role UIs, modern tooling.
Trade-off: Client-side rendering; live demo depends on backend availability and CORS configuration.
Node.js / Express REST API
Why: Single language across stack, rapid endpoint development, straightforward middleware for auth and roles.
Trade-off: Monolithic; all features share one process and deployment unit.
MongoDB + Mongoose
Why: Flexible document model fits evolving assessment schemas, child profiles, and nested category scores without rigid migrations.
Trade-off: Less rigid relational integrity; application-level consistency required for cross-document references.
JWT + role middleware
Why: Stateless authentication with explicit role checks (caregiver / expert / admin) for protected routes.
Trade-off: Token revocation and long-lived sessions need careful handling; OTP adds email dependency.
Groq API for AI features
Why: Fast inference, structured JSON responses, suitable free-tier for academic prototype. Used for quiz generation and Emotion Explorer content.
Trade-off: External dependency; rate limits, cost, and availability affect live demo.
Rule-based game recommendations
Why: Deterministic mapping from assessment category scores to therapy games is transparent, testable, and does not require an LLM call for every recommendation.
Trade-off: Less adaptive than a pure generative approach; mapping table must be maintained as games change.
6. Current Limitations (Honest Scope)
- Academic prototype — built and evaluated as a team Final Year Project, not a production clinical system.
- No clinical validation — assessments and AI-generated content have not undergone clinical trials or regulatory review. Not a diagnostic tool.
- Team project — contributions are collaborative; no single person owns the entire codebase.
- External service dependency — live demo relies on MongoDB, email provider, and Groq API.
- Frontend deployment only — public Vercel link is the frontend; full backend + database require local or private deployment.
- Limited automated testing evidence — core flows are demonstrated via seed scripts and manual presentation rather than comprehensive test suites.
- AI output quality — generated questions and scenarios depend on prompts and model behaviour; human review remains advisable.
7. Interface & Demo
Live frontend (may run with limited/mock data depending on external services):
Real screenshots and a short technical walkthrough video can be added here when available. No generated or placeholder images are used.
Medical & AI Disclaimer
AutiSmart is an academic and assistive software prototype. Its assessments, AI-generated recommendations, and screening-related features are not a substitute for professional clinical diagnosis, evaluation, or treatment. The project must not be interpreted as clinically validated or as a medical device.