Lumina PTE — AI-Powered SaaS PTE Learning Platform

SLOW, UNSCALABLE MANUAL MOCK TEST EVALUATION

Pearson Test of English (PTE) candidates require immediate, objective feedback on speaking and writing mock tests to prepare effectively for exam dates. Traditional coaching centers rely on human evaluators, resulting in long feedback cycles.

Operational Challenge:

Human evaluation was unable to scale during peak exam seasons, forcing students to wait 24 to 48 hours to receive scores on speaking fluency, pronunciation, and essay grammar.

Core Pain Points:

• 24+ Hour Evaluation Delays: Delayed feedback hindered continuous practice momentum for students.

• Subjective Scoring Variations: Human graders introduced inconsistent scoring criteria across test modules.

• High Operational Costs: Scaling human evaluators for thousands of daily test takers was financially unsustainable.

ASYNCHRONOUS AI SCORING AND MULTI-TENANT SAAS ARCHITECTURE

Designed an AI-powered SaaS platform using Next.js for the student frontend and Python evaluation microservices for speech-to-text, acoustic pronunciation analysis, and NLP essay grading.

System Architecture Highlights:

• Non-Blocking Queue Worker Pattern: Audio recordings submitted during mock tests are dispatched directly to BullMQ queues backed by Redis.

• Python NLP and Speech Workers: Specialized background workers evaluate phoneme accuracy, oral fluency, reading alignment, and grammatical syntax.

• Multi-Tenant SaaS Foundation: Built role-based access control for students, tutors, and institutional coaching centers with custom subdomains.

KEY TECHNICAL DECISIONS

1. Offloading Heavy AI Computation to Asynchronous Queues:

Processed all audio and writing submissions asynchronously instead of running evaluations inside HTTP request handlers to prevent timeout errors and allow independent queue auto-scaling.

2. PostgreSQL Row-Level Tenant Isolation:

Implemented tenant scoping at the Prisma ORM layer using middleware, ensuring strict data isolation between competing PTE coaching institutes sharing the infrastructure.

3. Stripe Subscription Webhooks and Usage Caps:

Built automated Stripe webhook handlers with exponential retry logic to process tier upgrades and monthly AI token allocations automatically.

BUSINESS IMPACT AND PLATFORM PERFORMANCE

• 15-Second Evaluation Feedback: Reduced test scoring turnaround time from 24 hours to just 15 seconds.

• 12,000+ Active Students: Successfully onboarded over 12,000 active test takers with a 99.92% platform uptime record.

• 98.4% Scoring Correlation: Achieved 98.4% correlation with official Pearson PTE metrics across 100,000+ evaluated mock tests.