VVS Academy / Pluto Academy

VVS needed to replace fragmented tutoring and academic-planning processes with one reliable platform. The system had to coordinate bookings, points, uploads, admission rules, and three user roles while preventing conflicts and keeping published guidance aligned with application results.

Built a Next.js and MongoDB platform with transactional booking and ledger workflows, server-projected admission datasets, direct-to-R2 uploads, role-aware dashboards, and four ChatGPA tools for prediction and planning. Domain tests and CI gates protect the workflows that carry the highest operational risk.

Junior Full-Stack Developer at VVS Education Pty Ltd. Architected core platform workflows, shipped the ChatGPA planning suite, established admission-rule contracts and domain testing, built the AI mathematics question-bank foundation, and implemented the CI/CD quality gate.

• Transactional tutoring bookings with double-booking prevention
• Auditable points-ledger workflows across student and admin operations
• Role-specific student, teacher, and administrator dashboards
• Four ChatGPA tools for GPA prediction, target-grade planning, Foundation admission matching, and UNSW application guidance
• Direct-to-R2 uploads without routing large files through the application server
• Admission-rule snapshots, matcher contracts, and allowlisted client projections
• AI mathematics question-bank pipeline derived from 210 pages, 8 units, 55 exercise sets, and approximately 95 reusable problem archetypes

• Next.js 16
• TypeScript
• MongoDB
• Vitest
• Cloudflare R2
• Vercel
• GitHub Actions
• LLM workflows

The mathematics pipeline turns curriculum analysis into reusable problem archetypes and a 300–500-question MVP. Deterministic grading and prerequisite-aware diagnostics remain programmatic, while the LLM layer is reserved for student-facing feedback so correctness-critical decisions do not depend on free-form model output.

Next.js 16 and TypeScript provide the application and server API surface, with MongoDB persistence, direct-to-Cloudflare-R2 uploads, and Vercel deployments. Server-side projections expose only allowlisted admission fields, rate limits protect planning APIs, transactional guards prevent double-booking, and domain tests validate rule snapshots, matcher contracts, and ledger behavior.

The highest-risk workflows combined concurrent booking changes, points accounting, changing admission requirements, and restricted datasets. The implementation needed to prevent schedule conflicts, preserve an audit trail, and ensure that published requirements and matcher output could not silently diverge.

Shipped a production platform across 55 page routes and 43 API routes, backed by 5,813 passing tests across 161 test files and a PR- and push-triggered CI/CD pipeline that validates lint, tests, and production builds before deployment.

Expand the mathematics MVP into adaptive practice paths with prerequisite-aware diagnostics, richer learning analytics, and production LLM feedback grounded in the structured problem-archetype library.