Abdullah Arshad

- Conceptualised and built the complete infrastructure, backend, and database schema for the Storyboard AI app from scratch (Python/Django + Node.js), with complex relational schemas for hierarchical media assets across multiple user roles. - Delivered v1 and v2 within 3 months (— two-column script export to PDF, ZIP archive generation, presentation-ready outputs, and a content planning system with APIs for nested content items. - Architected an extensible content management framework (diverse media types, attachments, validation rules) with metadata handling and relational mapping across channels, mediums, and asset types. - Built a multi-provider LLM content generation system (OpenAI GPT, Claude, Gemini, TogetherAI) using LangChain for automated article creation, categorisation, and SEO optimisation (— 90% content-quality accuracy). - Built a voice/audio generation system using the ElevenLabs API for multi-language voice-over generation, plus a content optimisation engine running automated A/B tests across LLM providers to tune quality, speed, and cost, with an analytics dashboard tracking provider performance and multi-format (PDF/HTML/Markdown) output. - Established GitLab CI/CD with Dockerised deployments for zero-downtime releases.

- Built a smart notification engine (Node.js/Python, RabbitMQ, cron) for real-time push notifications, plus email templates via Amazon SES; added testing, logging, error handling, and Swagger docs. - Applied machine learning to optimise notification timing and content personalisation based on user engagement patterns. - Created a custom Golang logger for backend/frontend; designed and optimised database tables for performance and scalability. - Refactored a monolith into modular architecture (services/controllers/routers) and delivered a full PostgreSQL-backed integration/unit test suite on GitLab CI.

- Designed and built LLMWithFailover, a cross-provider LLM abstraction layer with Redis-backed circuit breakers, exponential backoff, and health checks (— zero downtime across OpenAI, Claude, and Gemini outages. - Scaled bulk AI content generation from 50 to 1,000+ articles/batch: parallelised topical map processing, tuned Celery staggering (3s → 0.5s), added watchdog recovery tasks. - Reduced on-page audit failures from 24,000+ to near-zero (7 root causes, automated recovery, Slack alerts); cut AI-driven DKN generation time by 75% via parallelisation and isolated retry/backoff queues. - Owned Brand Vault, a RAG-powered enterprise knowledge base end-to-end: soft-delete/archive, audit logging, bulk parallel GCS uploads via Celery group tasks, vector-based retrieval. - Built 6+ CMS connectors (WordPress, Salesforce, Drupal, Ghost, Craft CMS, BigCommerce, Cloudflare Pages) with automated error recovery, slug-conflict detection, and hero image deduplication. - Implemented LLM cost tracking and Pyroscope profiling; redesigned Celery queue architecture, cutting scan overhead from 38 queues to single-queue lookups.

- Built a full payroll disbursement system in NestJS (RabbitMQ, Easypaisa API, automated audit reporting) (— reduced processing time by 60%; applied PostgreSQL optimisation and Redis caching for real-time calculations. - Wrote a cross-platform Node.js SDK for ZKTeco biometric integration (Linux/macOS/Windows) with automated background sync. - Designed a flexible policies/holidays configuration engine and custom scheduler, plus cron jobs automating timesheet validation (— backed by a full Jest unit test suite. - Built Python-based automated reconciliation (Pandas transaction matching, cutting manual processing time by 80%) and OpenAI-powered transaction analysis for anomaly detection and natural-language compliance summaries.