PLEASE READ

We have too many features to demo in 5 minutes time, so please check out the website below to try out yourself! ^.^

Try out our app LIVE at

https://dorachua.biz

One Page Proposal

https://acrobat.adobe.com/id/urn:aaid:sc:ap:69740e6e-3a70-4151-90c2-89f258e5f0f4

Slide Deck

https://acrobat.adobe.com/id/urn:aaid:sc:ap:5ab5ac3e-ccdb-4073-8fbf-6e5f5f5518e6

Inspiration

Our project aims to address the problem of fragmented digital experiences and disconnected support systems across campus life at SP.
Students and staff currently navigate multiple platforms for well-being support, knowledge resources, and productivity tools, leading to inefficiencies, information overload, and missed opportunities for connection.
Additionally, existing campus resources are often buried in lengthy policy documents or scattered across websites, making it difficult to find timely, accurate answers when they're needed most.

What Does our App Do?

SP SuPer App provides a unified solution through three integrated pillars:

Feature	Description
Stay Positive	Tools that promote well-being
Stay Prepared	RAG-powered academic assistance
Stay Productive	Productivity apps for students & staff

Stay Positive creates a culture of appreciation and emotional well-being.

Feature	Description
SP Heartbeat	Digital appreciation wall that promotes gratitude and positivity.
SP Confide	AI-powered anonymous journal supporting emotional well-being.
SP Starboard	Campus-wide voting platform for fun and engaging themes.
SP Photobooth	A social photobooth that encourages students and staff to share images and amplify positive engagement.

Stay Prepared leverages RAG (Retrieval-Augmented Generation) technology to deliver grounded, trustworthy answers from official SP content.

Feature	Description
SP Concierge	Campus Q and A powered by Retrieval-Augmented Generation that provides grounded answers from official SP content.
SP Policy Guide	Plain-English explanations of school policies using verified institutional documents as the source.
Course Advisor	Course exploration assistant enhanced by Word document ingestion, allowing course chairs to continuously add new knowledge.

Stay Productive streamlines workflows with various tools including several that are AI-powered.

Feature	Description
SP Scheduler	Poll-based meeting coordinator to collect availability and send invites
SP AutoGrader	AI-powered assignment feedback to help students improve their work.
SP Feedback Coach	Generates constructive peer feedback for academic and project tasks.
SP Green Agent	CarbonCode uses AI to calculate your software's carbon footprint and then suggests more efficient code.
SP Lab Assistant	AI-powered, voice-enabled lab data capture system that converts speech into structured lab records using dynamic test configurations and an agentic AI parser.

How We Built It

We built SP SuPer App using the following technologies:

Category	Technology
Frontend	Next.js, React, TypeScript, Tailwind CSS
Authentication	Google OAuth via NextAuth
APIs	Next.js API Routes
Database	Firebase Firestore
Storage	Firebase Storage
RAG / Search	Supabase with pgvector
Embeddings	Google Generative AI embedding zero zero one
Document Intake	Mammoth for Word document parsing
Email	MailerSend for transactional messages
Hosting / CI	Vercel (production: https://dorachua.biz)

We built SP SuPer App using Next.js 16.0.7 with Turbopack and TypeScript, leveraging the App Router for optimal performance and developer experience.

The frontend is styled with Tailwind CSS, while authentication is handled by NextAuth with Google OAuth.

For data persistence, we use Firebase Firestore to manage user profiles, meetings, advice journals, and polls.

The RAG (Retrieval-Augmented Generation) capabilities are powered by Supabase with pgvector for vector similarity search, combined with Google Generative AI's embedding-001 model to generate 768-dimensional embeddings from course content.

We implemented a custom Word document ingestion pipeline using Mammoth to parse .docx files, enabling course chairs to easily add new knowledge to the system.

Transactional emails (meeting invites, notifications) are sent via MailerSend with click tracking.

The entire application is deployed to Vercel with CI/CD integration, running in production at **[https://dorachua.biz]

We also implemented role-based access control with dedicated dashboards for students, staff, course chairs, and administrators to ensure appropriate feature access across the platform.

AI Lab Assistant (Stay Productive)

The AI Lab Assistant is implemented as a voice-first, AI-assisted workflow within the SuPer App ecosystem.

Voice input is captured in the browser and transcribed.
Transcripts are sent to an agentic AI parser that maps natural language into structured lab fields, based on dynamically configured lab schemas.
Test records automatically capture tester identity (from Google OAuth) and timestamps, ensuring auditability.
All lab data is stored in Firestore with real-time updates across clients.

The AI Lab Assistant is deployed on Vercel at: 👉 https://ai-lab-assistant-steel.vercel.app

Smart Paradigms (Stay Productive)

Smart Paradigms is integrated as an AI-powered grading and academic support module.

Uses Retrieval-Augmented Generation (RAG) with Pinecone to ground grading and Q&A responses in lecturer-provided materials (lecture notes, PDFs, rubrics).
Supports custom rubrics, report summarisation, and recommended scoring ranges.
Provides instant preliminary feedback for students while preserving lecturer oversight.
Fully serverless using Next.js API routes and deployed on Vercel.

Smart Paradigms is hosted at: 👉 https://smart-paradigms.vercel.app/

Both AI Lab Assistant and Smart Paradigms share the same SuPer App design philosophy, authentication patterns, and cloud-native deployment approach.

Challenges We Ran Into

Getting the emails to work -- DNS verification
Getting the RAG system to work accurately

Auth Environment Mismatch: Local OAuth kept redirecting to production; solved with getNextAuthUrl() and split
Firestore Index Errors: Composite queries for advice/polls failed until custom indexes were defined and deployed; required patience for asynchronous index build.
Email Delivery Roadblocks: MailerSend returned 401/422 due to API key rotation and unverified sender domain; mitigated via domain verification and trial sender.
Accurate Link Routing: Meeting invite links pointed to localhost or tracked domains; standardized vote links using getNextAuthUrl() and clarified MailerSend click tracking.
Display Name Consistency: Emails showed Google account name instead of custom displayName; fixed by prioritizing session.user.displayName.
RAG Ingestion from Word: Parsing .docx reliably and segmenting content for embeddings was tricky; implemented Mammoth-based pipeline and section-aware upsert logic.
Voice-to-Data Mapping: Ensuring AI Lab Assistant correctly maps spoken phrases to dynamic lab fields required prompt iteration, type validation, and fallback handling.
Keyword Search UX: Preventing duplicate content required search with safe ilike filters; added input sanitization to avoid %/_ wildcard pitfalls.
Role-Based Access: Ensuring coursechair/admin-only features needed clear Firestore role documents and guarded API routes.
Multi-Environment Deploys: Coordinating Vercel production with Firebase services demanded careful config, secrets management, and regression checks.

Accomplishments We Are Proud Of

We are proud of the button shadow hover effect, which turned out surprisingly delightful.
Building and integrating AI Lab Assistant, enabling hands-free, voice-driven lab data capture with structured, auditable outputs.
Delivering Smart Paradigms, which meaningfully reduces grading workload while improving feedback speed and consistency.
We also overcame technical and time-constraint challenges that pushed our skills further.

What We Learned

We learned how to code, how to use React, and how to structure a full project over a short timeframe.
How AI can augment—not replace—human decision-making, especially in education and laboratory workflows.
The importance of grounding AI outputs using RAG and schema-aware parsing to build trust and reliability.
How modern vibe coding and AI-assisted development can accelerate prototyping, debugging, and iteration under tight timelines.
We also gained experience in collaboration, troubleshooting, and adapting under pressure.

What Is Next for "EXAMPLE TEMPLATE SUBMISSION"

We hope future hackers can use this as a reference for writing clear and complete Devpost submissions.
We may continue building on the project, polishing features, or exploring new enhancements.