Dat Nguyen

AI Engineer with 2 years of experience in applied AI, specializing in production-grade LLM applications, multi-agent systems, and agentic RAG workflows. Hands-on experience building tool-calling agents, hybrid retrieval, knowledge graph-based RAG, LLM evaluation and observability pipelines, with a strong interest in self-hosted AI infrastructure. Author of 3 research papers in AI and data mining.

AI ENGINEER at BLUE DRAGON COMMUNICATIONS (2026-04 – Present)

T-Station AI / Hankook Tire Korea Conversational commerce chatbot for Hankook Tire Korea.

• Multi-Agent System: Designed and built a production FastAPI conversational AI service using BAML, with gpt-5.6-terra for orchestration and gpt-4.1-mini for Leading, Discovery, Transaction, and Support agents; deployed on AWS using Docker Swarm.
• Tool-Calling Flows: Integrated 40+ domain tools and backend API operations covering tire search, recommendations, vehicle compatibility, pricing, coupons, inventory, stores, reservations, orders, tracking, and warranty workflows.
• Korean FAQ RAG & Evaluation: Built an incremental FAQ ingestion and hybrid retrieval pipeline using Qdrant, text-embedding-3-large, BM25+ with kiwipiepy tokenization, and RRF; evaluated retrieval using Precision@K, Recall@K, and response generation using faithfulness and answer correctness.
• Agent Memory & Context Management: Implemented Redis-backed structured conversation state and context recovery for multi-turn vehicle, tire-size, product, quantity, store, reservation, and order workflows.
• Agent Evaluation & Prompt Optimization: Generated and curated synthetic evaluation sets labeled with expected agents and tool calls; measured routing accuracy and tool-selection accuracy, optimized prompts using DSPy, and monitored LLM workflows with Langfuse and Grafana.

AI ENGINEER at VIETNAM SILICON (2025-04 – 2026-04)

AI ENGINEER — PlantDoctor / AgriChatbot at VIETNAM SILICON (2025-08 – 2026-04)

A durian pest and disease diagnostic sub-agent within the AgriChat multi-agent system.

• Agentic RAG Architecture: Built a multimodal RAG workflow with LangGraph and BAML, using gemini-3-flash for Knowledge Graph construction, Qwen3-30B-A3B for query rewriting and response generation, and Serper API for conditional web fallback.
• Ontology-Guided Knowledge Graph Ingestion: Built a parsing, chunking, and ingestion pipeline using Graphiti to transform Thai durian pest and disease documents into a Neo4j Knowledge Graph; applied a domain ontology to normalize Thai terminology, enforce KG consistency, and improve entity resolution and retrieval quality.
• Thai Hybrid Retrieval: Built a low-latency hybrid retrieval engine using PyThaiNLP tokenization, Neo4j BM25/Lucene full-text search, and Milvus HNSW vector search with snowflake-arctic-embed-l-v2.0; reranked lexical and semantic results with RRF.
• Multimodal Visual Search: Built an image retrieval pipeline using DINOv3 and Milvus HNSW to identify durian pests and diseases from photos and retrieve related diagnostic and treatment knowledge from Neo4j.
• Evaluation, Deployment & Observability: Built retrieval and generation evaluation datasets; used Redis for context and caching, DSPy for prompt optimization, Docker on AWS for deployment, and Langfuse and Grafana for observability.

AI ENGINEER — Talent Market Intelligence System at VIETNAM SILICON (2025-04 – 2025-08)

AI-powered recruitment platform for automated CV-JD matching.

• Backend Architecture & Deployment: Built a scalable asynchronous CV-screening backend using FastAPI and RabbitMQ; containerized the system with Docker and deployed it on AWS.
• Information Extraction: Built a parsing pipeline using Docling and a self-hosted LLM (mistralai/Mistral-Small-3.1-24B-Instruct-2503) via BAML to extract structured JSON data from CVs and JDs with near 100% accuracy.
• Hybrid Candidate Retrieval: Built a PostgreSQL-based hybrid retrieval pipeline with semantic indexing (pgvector + HNSW) and keyword indexing (BM25/Lucene); used RRF and Cross-Encoder re-ranking to retrieve Top-K candidates.
• Advanced Scoring Engine: Built a configurable LLM-based scoring engine to rank candidates by job stability, education, skills, and job requirements.

DATA SCIENTIST INTERN at TRUSTING SOCIAL (2023-08 – 2024-07)

Model Experimentation & Data Labeling Supported deepfake research and training data preparation.

• Deepfake Model Experimentation: Reproduced and tested open-source deepfake models, built runnable demos, and summarized findings, setup requirements, and technical limitations for data scientists and research engineers.
• Dataset Preparation: Collected, cleaned, transformed, and annotated datasets for Computer Vision and Text-to-Speech model training and evaluation, using CVAT for image annotation.
• Quality Assurance: Audited annotations, resolved inconsistencies, tracked progress and quality metrics, and collaborated with the Data Science team to improve guidelines and QA workflows.

B.Sc. in Data Science – Ho Chi Minh City University of Technology (2021-08 – 2025-09)