AI Engineer - Agentic Workflows & MLOps

KhangNguyen AI Engineer

Thu Duc Ward, Ho Chi Minh City – Viet Nam

094******* · # ****************@*****.*** · § khang3004 · ï KhangDS Professional Summary

AI Engineer specializing in Agentic Workflows (LangGraph, MCP) and scalable Machine Learning pipelines. Currently pursuing a Master’s degree in Data Science at HCMUS, focusing on Knowledge Representation, NLP, and Machine Learning. Proven track record in orchestrating multi-agent systems, optimizing semantic search architectures, and deploying production-ready AI solutions via Docker and FastAPI. Education

University of Science (HCMUS) Ho Chi Minh City

Master of Science in Data Science 2025–Present

Focusing on Natural Language Processing, Knowledge Representation, and Machine Learning. University of Technology (HUTECH) Ho Chi Minh City Bachelor in Data Science, GPA: 3.7/4.0 2021–2025

Class President; Bronze Medalist at Vietnam Mathematical Olympiad 2023–2024. AI Vietnam Remote

AIO2023 Program 2023–2024

Intensive program on advanced AI/ML & Data Science. Experience

Vietnam Silicon Company Ho Chi Minh City

AI Engineer Intern May 2025–Aug 2025

Talent Market Intelligence & Conversational AI System

System Architecture & MLOps: Architected a highly scalable microservices platform using FastAPI, RabbitMQ, and Docker, reducing CV screening time by 75% and enabling batch-processing of 100+ CVs in parallel.

Agentic Workflow Orchestration: Designed a ReAct-based conversational agent utilizing LangGraph and Model Context Protocol (MCP) with 50+ integrated tools, fully automating complex HR database queries via natural language.

Advanced Search Pipeline: Engineered a hybrid semantic search engine combining pgvector (HNSW) and FAISS with Sentence Transformers, achieving 92% accuracy in skill matching with sub-3s pipeline latency.

Resilience & Scalability: Implemented 24-API key failover rotation for Gemini 2.5 Flash LLM, ensuring 99.5% system uptime and continuous real-time streaming capabilities. Selected Projects

AgentSQL-Asym: Stateful Asymmetric Text-to-SQL GitHub. . . . . . . . . . . . . . . . . . . . . . . . Overview: Stateful asymmetric multi-agent framework orchestrated as a LangGraph State Machine to optimize Execution Accuracy and API costs on the BIRD-SQL benchmark. Impact: Achieved SOTA 80.20% Execution Accuracy (EX) and 74.50% Valid Efficiency Score (VES) on the 500-sample BIRD Mini-Dev set, reaching 92.0% on simple tasks. Orchestrated a strict 2-candidate budget using Gemma-4-31B to cut token costs by 90% and reduce latency to 82.40s. Integrated an ephemeral SQLite sandbox for runtime query validation and a targeted resilient corrector running Gemini 2.5 Flash. Tech Stack: Python LangGraph Qdrant/FAISS SQLite Sandbox Docker Omni-IPS: Neuro-Symbolic AI & GraphRAG Solver GitHub. . . . . . . . . . . . . . . . . . . . . . . . . Overview: Production-ready Neuro-Symbolic and GraphRAG-powered expert system resolving complex problems in Chemistry, Plane Geometry, and Elementary Algebra. Impact: Developed a domain-agnostic deterministic reasoning engine (forward/backward chaining) ensuring 100% logical correctness. Integrated a model-agnostic structured semantic parser and a multi-tier Qdrant L2 1/2

verification pipeline to eliminate rule mapping hallucinations. Ingested Euclid’s axioms and Wikidata reactions into a bipartite Neo4j knowledge graph (127 rules, 243 facts). Engineered an async real-time explanation streaming engine with FastAPI and Streamlit to eliminate UI latency. Tech Stack: Python Neo4j Qdrant FastAPI Streamlit Docker EduMIND: Bilingual Lecture Assistant & Active Learning GitHub. . . . . . . . . . . . . . . . . . . . Overview: Enterprise-grade bilingual lecture assistant and human-in-the-loop active learning pipeline for code- mixed Vietnamese-English academic environments.

Impact: Deployed a bilingual speech note-taker using Whisper ASR and a teencode resolver. Formulated a mathematical token-level Code-Mixing Index (CMI) to compute language transition metrics. Engineered an

“Anti-Forget” RAG engine with layout-aware PDF chunking, Qdrant hybrid search, and cross-encoder re-ranking. Integrated an active learning loop via Label Studio and a Flask ML backend to dynamically update the vector database and archive gold-standard transcripts.

Tech Stack: Python Whisper ASR Qdrant Label Studio Streamlit Docker Booking.com Hotel Analytics GitHub. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview: Multi-modal prediction and segmentation pipeline for hotel reviews (text/image/metadata). Impact: Fused ResNet18 image features with tabular data for regression (R2 = 0.78) and built a stacking ensemble for quality classification (ROC-AUC = 0.89). Packaged via Docker for complete reproducibility. Tech Stack: PyTorch Scikit-learn K-Means/DBSCAN Docker Vietnamese Financial Sentiment Analysis GitHub. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview: Hybrid ML/DL modeling for Vietnamese stock-market headline sentiment. Impact: Deployed a full NLP pipeline (PhoBERT, BiLSTM) managing data via MongoDB, achieving an out-of-sample accuracy of 89.2%.

Tech Stack: PhoBERT XGBoost MongoDB Docker

Technical Skills

Languages & Frameworks: Python (Advanced) FastAPI Streamlit Java JavaScript R Swift Agentic AI & Orchestration: LangGraph LangChain LangSmith MCP A2A Protocol Groq Data Science & ML: PyTorch HuggingFace Sklearn ONNX vLLM OpenCV Pandas Databases & Data Eng: PostgreSQL Neo4j SQLite MongoDB Redis Chroma/Milvus Spark MLOps & Workflow: Docker MLflow Ray RabbitMQ Amazon S3 Linux Git Accomplishments & Languages

Certifications: DeepLearning.AI – CNN Specialization DataCamp – Data Scientist Professional Awards: 2x Bronze Medals – Vietnam Mathematical Olympiad (2023, 2024) Languages: English (IELTS 6.0 - Currently upgrading), Vietnamese (Native) 2/2

Contact this candidate