Nguyên (Wynn)

Built, at the CEO’s request, a full-stack prototype of an LLM-powered AI assistant (personal well-being and organization focus) ahead of the planned product launch, using Next.js, shadcn/ui, Hono.js and Vercel AI SDK with Google Gemini: streaming chat, tool/function calling and autonomous notification workflows.

Implemented intent detection to classify user requests (e.g. create an event, track an event) and route them to the matching tools, with automatic context switching based on recent conversation history and dynamic suggestions.

Prototyped retrieval-augmented generation (text chunking, pgvector) to ground answers in external knowledge sources.

Wrote the chatbot’s system prompts with the CO-STAR framework to keep responses grounded and consistent across AI workflows.

Built REST APIs with Hono.js, stored conversation history in MongoDB with Mongoose (Azure Cosmos DB), and implemented authentication with better-auth, including Microsoft SSO.

Set up CI/CD with tag-triggered GitHub Actions workflows that build Docker images, push to Azure Container Registry and deploy automatically to Azure Container Apps.

Built event and note management tools, with a domain specialist, that automatically categorize notes using a custom algorithm.

Completed structured courses in software engineering, machine learning and data analytics (Coursera, Zero To Mastery).

Built guided and independent projects in web development, Python and data analysis (GitHub, Kaggle).

Developed and maintained backend services with Java Spring Boot for enterprise sales reporting (daily invoices aggregated into monthly sales and profit reports).

Fixed excessive per-invoice ORM queries in the reporting module: the service loaded each daily invoice entity separately through the ORM, and report exports by ~5 users nearly brought down the company's 8 GB Ubuntu server. Moved aggregation into the database with optimized queries and stored procedures; in local testing with a month of ~100 invoices (1-5 items each), export time dropped from about 10 minutes to about 30 seconds.

Joined a Spring Cloud service-oriented backend project that included a Spring WebSocket notification service. When an authentication issue appeared only on staging and the source was missing from GitLab (local and staging had diverged), recovered the deployed code by reverse engineering to reproduce and investigate it.

Set up a spare machine as a Docker-based test server to reproduce staging conditions; worked with CI/CD pipelines and Ubuntu Server.

Helped interns by explaining OOP design patterns and RESTful API standards.

Joined as a Java fresher (2-3 months of training), then moved to a Tier-1 automotive client project as a test engineer, in the second wave of engineers after the original team had left.

Wrote automated unit tests in C++ with Cantata for AUTOSAR-compliant embedded software, including hard-to-build modules assigned across the codebase.

Reviewed teammates' code and collaborated with internal and international teams across Europe, Russia, China and India.

Used Jira for ticket management and Jenkins to build sandbox environments.