Wais | Devpost

About the Project Inspiration For the InnOlympics 2026 "pagkatuto" (education) track, we wanted to bridge the digital divide in the Philippines. Many students and schools face unstable or non-existent internet connections. We were inspired to create a truly offline-first learning environment that still feels modern, premium, and capable of delivering personalized education. WAIS (formerly LearningHub PH) was born to empower Grade 5 classrooms with a seamless, tactile, and intelligent learning experience, regardless of internet connectivity.

What it does WAIS is a dual-role (Student and Teacher) offline-first Android application.

For Students: It provides a rich, distraction-free environment to read lesson modules (Science, Math, English) and take quizzes. Progress, mastery metrics, and quiz results are securely stored entirely on the device. For Teachers: It acts as a pocket classroom manager. Teachers can view record books, track individual student performance, and generate complete, curriculum-grounded lesson plans and quizzes using our AI assistant, GuroBot. The Offline Magic: Devices synchronize data entirely offline using QR codes! Students can export their profiles or quiz results to a QR code, which the teacher scans to update their central record book. Teachers can generate lessons and share them back via QR for students to ingest. How we built it We utilized a fully native Android stack written in Kotlin, focusing on resilience and local processing.

UI & Design: We used Jetpack Compose and adopted the "Scholar Indigo" design system (generated via Google Stitch). It emphasizes an "Organic Editorialism" aesthetic with tactile, paper-like surfaces, accessible 48dp touch targets, and playful yet professional accents. Local Persistence: Room Database acts as the single source of truth for modules, student records, and quiz attempts. Offline Sync: We integrated CameraX, Google ML Kit for barcode scanning, and ZXing to encode/decode JSON payloads into high-density QR codes for peer-to-peer data transfer. AI Integration: We built a custom CurriculumStore using a native, in-memory Kotlin vector index. We use the cosine similarity formula to match student needs with the curriculum: $$ \text{similarity} = \cos(\theta) = \frac{\mathbf{A} \cdot \mathbf{B}}{|\mathbf{A}| |\mathbf{B}|} $$ When online, GuroBot orchestrates the Gemini 1.5 Pro API via native HTTP REST calls to generate grounded lesson plans and contextual quizzes, which are then cached offline. Challenges we ran into Offline Peer-to-Peer Data Transfer: Ensuring complex JSON data (like full quiz results, metrics, and lesson modules) could fit within the data capacity limits of a QR code while maintaining high scannability. We had to heavily optimize and compress our data payloads. AI Hallucinations & Constraints: We had to implement strict grounding policies for Gemini. A lesson plan or quiz is only generated if the retrieved curriculum chunks surpass a strict similarity threshold (e.g., $\cos(\theta) > 0.72$). We also built fallback offline logic to provide teaching assessments when the API is completely unreachable. UI/UX for Grade 5: Balancing a premium design with the accessibility required for 10-11 year olds. We carefully applied the Plus Jakarta Sans and Inter typefaces to maintain extreme legibility without losing the "Scholar Indigo" identity. Accomplishments that we're proud of Successfully implementing a completely offline data synchronization system using QR codes, bypassing the need for cloud infrastructure in the classroom. Building a native Kotlin RAG (Retrieval-Augmented Generation) pipeline from scratch—featuring text chunking, local vector storage, and embedding calculations—without relying on heavy third-party vector databases. Translating our high-fidelity Stitch UI mockups perfectly into a responsive, production-ready Jetpack Compose codebase. What we learned Data Compression is Key: We learned a massive amount about payload optimization when relying on visual data transfer (QR codes). Every byte counts. On-Device Vector Math: Implementing cosine similarity locally taught us how efficient mobile devices have become for lightweight machine learning tasks and chunk retrieval. Design Systems: Strict adherence to a design system significantly speeds up UI development and creates a remarkably cohesive user experience. What's next for Wais Automated Grading via ML: Using local ML Kit models to optically grade handwritten short-answer quizzes. Broader Curriculum Support: Expanding the CurriculumStore beyond Grade 5 to support the entire K-12 MATATAG curriculum. Mesh Networking: Exploring Wi-Fi Direct or Bluetooth mesh networking as an alternative to QR codes for faster, seamless offline sync across an entire classroom.