Distributed systems and backend platforms built for measurable production impact.

// Context & Problem

Multi-tenant omnichannel POS & ERP platform operating across 200+ retail stores with real-time checkout, distributed inventory management, and automated tax e-invoicing.

// Core Engineering Contributions

• › Engineered core Golang microservices for payment processing, stock locking, B2B workflows, and automated per-order tax filing via Viettel/MISA APIs.
• › Architected an asynchronous event-driven messaging layer using Azure Event Hubs (Kafka protocol) paired with the Transactional Outbox pattern.
• › Standardized GitOps CI/CD on Kubernetes (AKS) using ArgoCD App-of-Apps, featuring automated pre-upgrade hooks running Goose DB migrations before pod rollout.
• › Integrated receipt and document intelligence using Mistral OCR on Azure AI Foundry and a Multi-Agent automated reporting system with Azure GPT-4o.

// Measurable Impact

• › Sustained reliable, real-time data synchronization across 200+ active retail stores.
• › Eliminated event loss and state drift between PostgreSQL and message queues via the Transactional Outbox pattern.
• › Maintained p99 latency under 85ms during peak shopping periods, verified through rigorous k6 stress testing.
• › Reduced manual deployment overhead by 85% and eliminated schema drift incidents through automated ArgoCD hooks.

// Context & Problem

Enterprise LMS platform serving tens of thousands of university students and lecturers, featuring custom assessment workflows and local AI-assisted tutoring.

// Core Engineering Contributions

• › Developed and maintained core Moodle plugins in PHP and JavaScript (OUStudio, Sketch, Assignment).
• › Authored comprehensive automated test suites using PHPUnit (unit testing) and Behat (behavior-driven acceptance testing).
• › Integrated a privacy-compliant local LLM gateway using Ollama AI to summarize quizzes and exams for students and faculty.
• › Implemented a multi-tier caching architecture with Redis Cache for high-frequency AI inferences and repetitive query paths.

// Measurable Impact

• › Reduced database query pressure by ~40% on critical learner endpoints during semester exams.
• › Significantly lowered compute overhead and token costs through intelligent response caching in Redis.
• › Maintained 100% data privacy compliance by hosting and executing all AI models on internal enterprise infrastructure.

// Context & Problem

Comprehensive internal IT asset lifecycle management system with cloud infrastructure as code (Terraform) and automated CI/CD deployment on Microsoft Azure.

// Core Engineering Contributions

• › Engineered type-safe REST APIs using NestJS and Prisma ORM backed by Azure Database for PostgreSQL.
• › Codified cloud infrastructure using Terraform (IaC) to provision Azure App Service, Azure Container Registry, and secure networking.
• › Constructed automated CI/CD pipelines in Azure Pipelines covering linting, unit testing, container build, and deployment.
• › Collaborated on an administrative dashboard in ReactJS featuring role-based access control (RBAC).

// Measurable Impact

• › Digitized 100% of employee hardware lifecycle events from onboarding provisioning to asset retirement.
• › Reduced feature deployment lead times from days to under 15 minutes with automated Azure Pipelines.
• › Recognized as an outstanding capstone project in the Rookie to Engineer program (Batch 7) and adopted into live operational use.

// Context & Problem

Real-time meal recognition and nutritional breakdown platform powered by Gemini 2.5 Pro on Google Vertex AI, semantic vector search with Qdrant, and serverless Azure Functions.

// Core Engineering Contributions

• › Architected serverless backend microservices using Azure Functions with TypeScript.
• › Integrated multimodal vision inference via Gemini 2.5 Pro on Google Vertex AI for granular food ingredient detection.
• › Implemented semantic similarity search using Qdrant Vector DB to normalize dish names and resolve abbreviations against nutritional databases.
• › Enforced edge security with Azure API Management (APIM) rate limiting and Azure Entra ID (Single Sign-On).

// Measurable Impact

• › Delivered complete image-to-nutrition analysis payloads in under 2 seconds.
• › Scaled elastically to thousands of concurrent requests with zero idle server overhead.
• › Automated client SDK generation and contract verification directly from OpenAPI definitions.