NGUYỄN HỒNG QUÂN

Sensei Assistant (Primary Focus)

• Built a GPT‑4 and GPT‑5 powered assistant with Retrieval‑Augmented Generation (RAG) to help engineers find and understand technical documents faster.
• Architecture: Azure OpenAI (GPT‑4), Azure Document Intelligence (OCR + layout), Azure Cognitive Search (hybrid semantic + vector), LangChain, React + TypeScript.
• Streaming (SSE): Server‑Sent Events with heartbeats, backoff/retry, graceful interruption.
• Redis: Session state store, idempotency keys, per‑user rate limiting, pub/sub for cross‑instance events.
• Markdown Renderer: Safe, readable Markdown with sanitization, fenced code blocks, syntax highlighting, inline citations and anchors.
• RAG Design:
  • Ingestion: PDFs/images → Azure Document Intelligence → normalized chunks with metadata.
  • Chunking: Section‑aware based on headings and layout; tables and captions handled.
  • Embeddings & Search: Azure OpenAI embeddings; Cognitive Search with vectors; hybrid scoring (BM25 + vector).
  • Re‑ranking & Citations: Re‑rank top‑k, add citation snippets with page anchors; de‑duplicate near‑identical chunks.
• Prompting & Guardrails: Grounded answers with citations, token limits, truncation rules, safe‑completion fallbacks; user‑friendly error handling.
• Reliability & Observability: App Insights tracing/logging with correlation IDs; latency metrics for ingestion, retrieval, generation.
• CI/CD & Quality: Repo bootstrap, lint/format/test scripts, Azure DevOps pipelines; secrets via Azure Key Vault.
• Outcomes: Reduced time to look up answers; supported Excel/PowerPoint/PDF/Word; ~1000 DAU.
• Role: First engineer across frontend, backend, AI workflows, and DevOps.

Bing‑style Document Search

• UI with facets, filters, preview panes for quick discovery.
• Pipeline: OCR/structure via Document Intelligence → normalization → Cognitive Search indexing (content + metadata + vectors).
• UX: Suggestions, filters by asset/type/date, highlight snippets, quick preview with page anchors, keyboard navigation.

MoC (Management of Change) Workflow for FPSO

• Workflows with approvals, audit trails; attachments with metadata/versioning.
• Role‑based access, notifications, links to supporting documents.
• Outcomes: Lead time per change reduced; improved compliance and audit readiness.

Core Tech: React.js, TypeScript; Node.js; Azure OpenAI, Document Intelligence, Cognitive Search; LangChain; Redis; Azure App Service/Functions; Azure DevOps; App Insights; Azure Key Vault.

• Built Edubao.org, a dynamic web platform for insurance packages targeting students and workers in Germany.
• Integrated a CMS‑based blog with Squidex.io to support dynamic content.
• Designed SEO‑friendly pages and enhanced website visibility.
• Created a search engine for German study programs using Elasticsearch.
• Automated deployments with Azure DevOps and later migrated to AWS.
• Technologies: Next.js, Gatsby, GraphQL, React.js, Elasticsearch, Azure DevOps, AWS Lambda.

• Developed transaction management dashboards for Heidelpay to support real‑time revenue tracking.
• Introduced dynamic filtering and chart visualizations for large datasets.
• Learned PHP to implement backend workflows processing 200+ TPS.
• Technologies: React.js, Symfony, MySQL, Chart.js.

• Migrated legacy AngularJS applications to React.js, improving maintainability and scalability.
• Built authentication middleware using Ruby and scalable internal services with Go.
• Technologies: AngularJS, React.js, Ruby, Go, Redis, RabbitMQ.

• Contributed to Golfconnect24.com, enabling real‑time online booking and scorecard management for golfers.
• Technologies: AngularJS.