Building reliable backend systems and practical AI-powered applications.
📍 Ho Chi Minh City, Vietnam · 🎓 Information Technology, UIT – VNU-HCM
💼 Open to Backend Developer / AI Engineer Intern & Fresher opportunities
I'm an Information Technology student focused on Backend Development and AI Engineering. I enjoy building backend services, data pipelines, and intelligent applications that integrate machine learning and LLM capabilities into real-world systems.
- Backend Engineering: REST APIs, authentication and authorization, database design, asynchronous processing, and service integration.
- AI Engineering: LLM applications, RAG, multi-agent workflows, NLP model evaluation, OCR, and AI service integration.
- Infrastructure: Event-driven processing, caching, Docker, and CI/CD pipelines.
- Currently Building: An enterprise AI agent platform focused on modular execution, tool governance, and human-in-the-loop workflows.
| Area | Technologies |
|---|---|
| Languages | TypeScript, JavaScript, Python, PHP |
| Backend | NestJS, Node.js, Express.js, FastAPI, Laravel |
| AI / LLM | PyTorch, LangGraph, RAG, BGE-M3, BM25, PaddleOCR |
| Data & Messaging | PostgreSQL, MongoDB, Redis, Qdrant, Apache Kafka |
| Frontend | Next.js, React, Tailwind CSS |
| DevOps & Tools | Docker, Git, GitHub Actions, Postman, MinIO / S3 |
Enterprise AI Agent Platform · In Active Development
A reusable platform that enables enterprises to build and manage secure AI agents across business workflows without developing separate agent infrastructure for each use case.
- Agent Architecture: Designed a modular, multi-tenant architecture separating the Control API, LangGraph-based Agent Runtime, background workers, and Studio Web.
- Execution & State: Developed agent execution foundations with REST APIs, SSE streaming, persistent run state, idempotent requests, and retry/recovery mechanisms.
- Tool Governance: Designing a secure Tool Gateway for MCP integrations, schema validation, permission enforcement, human-in-the-loop approvals, and audit logging.
- Security & Isolation: Developing authentication, RBAC, and workspace isolation to enforce secure access across enterprise tenants.
- RAG Infrastructure: Configured Qdrant vector search with multilingual E5 embeddings, Redis, and MinIO using Docker Compose.
Stack: Python · FastAPI · LangGraph · PostgreSQL · Redis · Qdrant · MinIO · Next.js · Docker
An AI-powered health assistant that interprets InBody measurements and retrieves evidence from health documents to support personalized nutrition and workout guidance.
- Multi-Agent System: Built LangGraph-based workflows with specialized agents for InBody analysis, nutrition planning, workout recommendations, and knowledge retrieval.
- Document ETL & OCR: Developed document-processing pipelines with PaddleOCR, text normalization, table-aware chunking, and parent-child retrieval to preserve document context.
- Hybrid RAG: Combined BGE-M3 embeddings (1024 dimensions), Qdrant vector search, and BM25 lexical retrieval for evidence-based question answering.
- Safety & Validation: Implemented deterministic safety checks, citation verification, and evidence-based response validation to reduce unsupported medical claims.
- Testing: Reported 152 passing automated tests and indexed 4,405 document chunks for retrieval.
Stack: Python · FastAPI · LangGraph · Qdrant · BGE-M3 · BM25 · PaddleOCR · PostgreSQL · Next.js · Docker
Vietnamese Toxic Comment Classification · Team Project
An NLP-based content moderation system combining Vietnamese social media data collection, hierarchical toxicity classification, and real-time comment monitoring.
- Data ETL & Annotation: Collected and preprocessed 48,670 Vietnamese comments from Facebook, YouTube, TikTok, VOZ, and Threads using multi-stage cleaning and LLM-assisted labeling.
- Model Evaluation: Benchmarked TextCNN, PhoBERT, ViDeBERTa, and XLM-RoBERTa. PhoBERT achieved 91.76% accuracy and 0.7943 Macro-F1 for binary toxicity classification.
- Hierarchical Classification: Evaluated explicit versus implicit toxic expressions, with XLM-RoBERTa achieving 83.79% accuracy on the second-stage classification task.
- Real-Time Processing: Integrated Apache Kafka, Python/PyTorch inference workers, Redis, Node.js APIs, and SSE streaming for live comment analysis and dashboard updates.
Stack: Python · PyTorch · PhoBERT · XLM-RoBERTa · Node.js · TypeScript · Apache Kafka · MongoDB · Redis · Next.js
MLOps Pipeline · Learning Project
A hands-on machine learning engineering project exploring reproducible experiments, dataset versioning, automated quality checks, and model delivery.
- ML Experimentation: Worked with the UCI Adult Income dataset to explore model training, hyperparameter tracking, and evaluation using MLflow.
- Data Versioning: Practiced reproducible dataset management with DVC and cloud object storage.
- CI/CD & Model Serving: Explored GitHub Actions workflows for automated testing, training, F1-score quality gates, and FastAPI-based model serving.
Stack: Python · Scikit-learn · MLflow · DVC · GitHub Actions · FastAPI
A food delivery platform integrating restaurant management, online ordering, digital payments, computer vision, and conversational AI to simplify food discovery and ordering.
- Backend Architecture: Developed modular NestJS backend services for restaurant management, product catalogs, order processing, authentication, and role-based access control.
- AI Food Recognition: Integrated a computer vision pipeline combining YOLOv8 food detection and EfficientNet-B2 classification to recognize 30 Vietnamese dishes from images.
- Conversational Ordering: Built a FastAPI-based AI chatbot service to interpret natural-language food requests and support order creation.
- Payments & Reliability: Integrated payment workflows with MoMo and VNPay, using transaction validation, idempotency, and order-state management.
- Infrastructure: Used PostgreSQL, Redis, BullMQ background workers, and Docker Compose to support asynchronous processing and service integration.
Stack: NestJS · TypeScript · Next.js · FastAPI · YOLOv8 · EfficientNet-B2 · PostgreSQL · Redis · BullMQ · Docker
An online examination and assessment platform designed to help educational institutions manage question banks, organize exams, evaluate students, and automate exam preparation.
- REST API & RBAC: Developed Laravel-based REST APIs for user management, academic subjects, question banks, and examination workflows with Sanctum authentication and role-based authorization.
- Exam Management: Built workflows for exam scheduling, student assignments, question distribution, automated scoring, and result management.
- Reliable Exam Submission: Implemented debounced answer autosaving, idempotent updates, and database transactions with pessimistic locking to prevent duplicate submissions.
- AI-Assisted Exam Planning: Developed an ExamOps AI Copilot architecture using FastAPI and LangGraph to support question-blueprint generation, deterministic validation, and human approval workflows.
Stack: PHP · Laravel · Next.js · TypeScript · FastAPI · LangGraph · PostgreSQL · Redis · Docker
- Backend Systems: API design, authentication, clean architecture, event-driven processing, distributed systems, and reliability engineering.
- Agentic AI & LLMs: Agent orchestration, RAG evaluation, context engineering, tool calling, MCP, AI safety, and observability.
- MLOps: Reproducible experiments, model deployment, automated evaluation, and CI/CD for ML systems.
I'm open to Backend Developer and AI Engineer Intern / Fresher opportunities, as well as collaboration on practical AI and open-source projects.
- GitHub: github.com/duonyyy
- Email: duongtungduong34@gmail.com
Build reliable systems. Make AI useful.


