Quek Xin Ying
Applied AI & Analytics student at Nanyang Polytechnic specializing in full-stack web applications, real-time socket networking, binary Protobuf protocol engineering, and predictive machine learning pipelines.
School Work & Academic Projects
View All School Work →During Singapore Chinese Chamber of Commerce & Industry (SCCCI) overseas trade missions, tracking 30–100+ business delegates across moving tour buses with paper manifests caused transit delays, untracked absences, and zero live cross-staff coordination.
Engineered real-time facial recognition verification (Face-API.js) with offline-first QR fallback, backed by atomic PostgreSQL transactional route allocations to eliminate double-assigned seats and Socket.io live sync across all officers' mobile devices.
Senior citizens experience acute digital exclusion and social isolation, while youth volunteers lack a structured platform for pairing. Existing portals fail elderly accessibility standards with rigid font scaling and poor contrast ratios.
Implemented an object-oriented scoring engine matching seniors with youth based on dialect affinity (Hokkien, Cantonese, Teochew), geographic proximity, and digital coaching goals. Features a root-rem font scaler, WCAG 2.1 AAA high-contrast modes, and Singpass-style verification.
Student workshop schedules, hackathons, and technical sub-chapters (Dev Swift, NYP Cloud, LIT) were announced across disconnected chat groups, resulting in lost registrations and poor student discovery.
Built a unified, mobile-first responsive hub with structured categorization for club tracks, interactive workshop timetables, and strict client-side form validation conforming to IBM SkillsBuild Web Design principles.
Side Projects & Open-Source Engineering
View All Side Projects →Distributed stream monitors and telemetry services frequently crash or become unmonitored zombies under network churn, lacking centralized process supervision and automated notification bridges.
Configured PM2 process orchestration with exponential restart backoff throttling to prevent CPU thrashing during network disruptions, combined with real-time Telegram alert bridges dispatching streamer online status transitions.
Social media operators managing high-cadence accounts face repetitive manual posting workflows, unexpected cookie expirations, aggressive platform rate limits, and frequent failures when uploading large multi-part media.
Created an interactive readline CLI and automation pipeline supporting dry-run simulation, chunked multi-part image and video uploads, exponential-backoff cookie rotation, and persistent metadata caching via MongoDB and serverless functions.
High-concurrency live streaming broadcasts emit unpredictable binary WebSocket data that silently drop during network spikes, while creators lack real-time telemetry and predictive models to forecast gift income or detect donor churn.
Engineered a fault-tolerant ingestion daemon with a heartbeat watchdog thread cycling pseudo-alive connections with zero downtime. Built an integrated Auto-ARIMAX time-series model predicting gift velocity and user lifetime value (LTV).
Live broadcast interaction telemetry is delivered over proprietary, undocumented binary Protobuf payloads, preventing third-party developers and researchers from building external integrations or accessibility tools.
Reverse-engineered binary wire traffic and nested varint field tags into complete Protobuf schemas, coupled with a sanitized offline testbed and automated multi-language build scripts outputting Python, Go, and TypeScript stubs.