I am an AI Engineer focused on building end-to-end AI systems involving Large Language Models, Speech AI, AI Agents, RAG, and GraphRAG.
My experience spans model fine-tuning, AI evaluation, system architecture, retrieval pipelines, and production-oriented AI services. I am particularly interested in understanding not only which model performs best, but also why a specific architecture should be selected under accuracy, latency, cost, and reliability constraints.
- 🔭 Current Focus: LLM Systems, Agentic RAG, Speech AI, and AI Evaluation.
- 🧠 Interests: LLM Safety, GraphRAG, Model Routing, Retrieval Systems, and Production AI.
- ⚙️ Engineering: End-to-end AI pipelines, backend services, asynchronous processing, and MLOps.
- 🎓 GPA: 3.45/4.0 | English: TOEIC 650.
- Posts and Telecommunications Institute of Technology (PTIT)
Information Technology — Minor in Information Systems
Aug 2021 - Jan 2026
AI Engineer | May 2026 - Present
- Built an LLM guardrail system for Jailbreak and Prompt Injection detection; fine-tuned Qwen3Guard with QLoRA and developed a hidden-state-based router, improving internal benchmark accuracy by 15.7% and Jailbreak recall by 6.1%.
- Designed an end-to-end Vietnamese AI Call Analysis system covering ASR, speaker/role attribution, PII masking, ITN, and structured outputs for downstream analysis.
- Built an evaluation framework across dozens of real-world calls using WER/CER, speaker-aware metrics, P95 latency, RTF, and accuracy-latency-cost trade-offs.
- Designed a Hybrid PII pipeline using deterministic rules, evidence-based routing, and selective V-LLM verification.
- Built Hybrid ITN and role-classification pipelines with rule-first processing, V-LLM fallback, batching, validation, and safe degradation.
Backend Developer | May 2025 - Aug 2025
- Built RESTful APIs using Spring Boot and Oracle Database following OpenAPI standards.
- Implemented Redis-based API rate limiting for traffic control.
- Refactored backend services into reusable components to improve maintainability.
| Category | Technologies |
|---|---|
| Languages |    |
| AI / ML |     |
| LLM / RAG |    RAG GraphRAG QLoRA LLM Evaluation |
| Backend |    Celery gRPC WebSocket |
| Databases |   Milvus Elasticsearch |
| MLOps / Cloud |    GitHub Actions DVC EC2 ECR |
| Project | Role & Impact |
|---|---|
| 🤖 IT Support Agent Platform (Sep 2026 - Oct 2026) |
• Built a stateful AI IT support platform using OpenAI Agents SDK and four-layer memory. • Engineered Hybrid RAG with pgvector, BM25, weighted RRF, reranking, and citation validation. • Achieved 0.7318 nDCG@10 on BEIR SciFact and 0.7860 nDCG@10 on adapted LoCoMo retrieval benchmarks. • Integrated MCP tools with access control, human approval, and idempotent ticket creation. • Implemented durable agent execution using Celery, Redis, and PostgreSQL. |
| 🧠 InsightAI: Agentic AI Learning Platform (Jun 2026 - Jul 2026) |
• Built an AI learning platform that converts videos, PDFs, and audio into lessons, summaries, quizzes, and mind maps. • Designed LangGraph Agentic RAG with intent routing, context grading, and web search fallback. • Implemented Neo4j GraphRAG and hybrid retrieval using Milvus and BM25. • Developed asynchronous AI content pipelines with Celery, Redis, and Whisper. • 🎥 Demo |
| 🥗 MyHealth: AI Nutrition & Meal Planning Platform (Oct 2025 - Dec 2025) |
• Built an AI nutrition platform for personalized meal planning and dietary consultation. • Designed Agentic RAG using LangGraph. • Fine-tuned BGE-M3 using QLoRA and contrastive learning, integrating embeddings with Elasticsearch. • Developed an SLSQP-based optimizer for personalized meal planning. |
- Artificial Intelligence with Python — Samsung (Jul 2024 - Aug 2024)
- Applications of Algorithm — Samsung (Jul 2023 - Aug 2023)