9 years — from software development to AI-model research to AI engineering in production — as technical leader and hands-on engineer — currently within Vietnam's first nationwide banking AI transformation with RemittanceAI and an AML Agentic System.
Multi-Agent SystemsBanking & Financial AILLM Fine-Tuning & EvalsResearch to Production
0K+AML Entities Screened
0M+Tokens / Day
0%RAG Groundedness
0×Eng. Throughput
0Years Experience
0Publications
01
TL;DR
🏦
Banking AI in Production
RemittanceAI + agentic AML live at Techcombank — processing cut from 17.5 to 3 min per transaction (−83%) at up to 1,200 document sets/day; 6/6 contractual AI KPIs passed, with 96.5% decision accuracy and every AML flag citing the regulation
🚀
LLM Serving Platform
200M+ tokens/day shared by RemittanceAI, Dora, and OneHiring — LiteLLM gateway over self-hosted vLLM, SGLang, and Triton on 5×H200 and 6×AMD MI355X, 500K+ requests/week at ~2s median, token spend cut 30%
🎯
Multi-Agent Platform
Dora — 5 agent roles on one LangGraph runtime, plan-and-execute checked by deterministic laws, MCP tools and human-in-the-loop; 96.6% auto-fill precision on public benchmarks, SFT + GRPO distilled into a self-hosted model
🧭
Platform Architecture
Reference architecture for OneHiring — 4 agents source-to-hire on Qdrant hybrid RAG at 98% groundedness, tenant-scoped with no cross-domain leakage; the blueprint 6 engineers build against
02
Experience
2026 – NowStaff AI Engineer & Program Manager
OneMount – AI Platform OneNexus
2025 – 2026Research Scientist
Physical AI Lab, Korea Evaluation Institute of Industrial Technology
2020 – 2025Research Scientist
Machine Vision Lab, National Research Foundation of Korea
2018 – 2020Sr. SW Engineer
Robert Bosch GmbH
2017 – 2018SW Engineer
TMA Solutions
Jan 2026 – PresentCurrent
OneMount
Staff AI Engineer & Program Manager – AI Platform OneNexus
Hanoi, Vietnam
Technical lead and program manager for AI on OneMount’s banking platform: sets direction with the CTO and engineering leadership across 6 domains, plans the AI project portfolio and who staffs it, and owns the architecture other teams build on. Took RemittanceAI and agentic AML to Techcombank production — processing cut from 17.5 to 3 min per transaction (−83%), 6/6 contractual AI KPIs passed — on an LLM serving platform architected to carry 200M+ tokens/day. Leads the group’s recommendation foundation model from research to production, and set the technical direction for Dora, where evidence-citing agents and deterministic rules replace a manual spec-to-pipeline loop. Across teams: the OneHiring reference architecture 6 engineers build against, and the build-vs-buy calls — agentic, RAG, or fine-tuning; self-hosted or metered API — decided on evals, latency, and cost.
Technical LeadershipProgram ManagementMulti-AgentLangGraphSFT · GRPORecommender SystemsLiteLLMvLLMMCP / A2ARAGQdrantKubernetesBanking AI
Jan 2025 – Jan 2026Research
Physical AI Lab
Research Scientist – Physical AI & Model Optimization
Seoul, South Korea
Research Scientist advising hospital and healthcare partners on privacy-preserving federated learning and embodied-AI adoption. Cut inference cost ~2× training-free via sparse MoE expert routing, quantized vision models to INT8/INT4 on Jetson Orin NX (8× smaller, <1% accuracy loss), and delivered 3× classification accuracy over CLIP on real-world agricultural imagery.
Federated LearningQuantizationTensorRTJetson OrinCUDAROS 2 · Isaac SimVLACLIP
Sep 2020 – Jan 2025Research
Machine Vision Lab
Research Scientist – Computer Vision and Multimodal AI
Seoul, South Korea
Research Scientist in computer vision and multimodal AI. First-author open-vocabulary segmentation +22.1 mAP over prior SOTA with zero manual annotation via a VLM knowledge-distillation pipeline; PAH-Net (Neurocomputing Q1) set SOTA on three clothes-changing re-ID benchmarks at 107 FPS and 32.6M parameters; delivered a real-time sports-broadcast detector exceeding contractual KPIs for a government-industry consortium.
Senior Software Engineer architecting enterprise B2B systems from scratch to production — Java Spring Boot/Spring Cloud microservices, Oracle/MongoDB, event-driven Kafka. Owned authentication (JWT), load balancing, and Jenkins CI/CD for a global HR portal serving 10,000+ users, and architected the company-wide social network for a ~400,000-employee organization. Awarded Best Performance 3×.
Java Spring BootSpring CloudOracleMongoDBKafkaRedisKubernetesJenkins CI/CD
Jun 2017 – Aug 2018Industry
TMA Solutions
Software Engineer
Ho Chi Minh City, Vietnam
Software Engineer delivering full-cycle Angular + Java production applications — the disciplined full-stack foundation that underpins every system shipped since.
AI coach on a national program of 22,000 students (QS Reimagine Education Awards 2026), personally coaching 80 mentees. Co-mentored the Security cohort with the VinSOC Head of AI Solutions and lead the Banking & Fintech cohort as main coach, guiding teams building multi-agent systems. Appointed a Committee Member of AI in Action, Vingroup’s AI community, setting the program’s technical review bar.
Appointed one of 4 company-wide AI Experts — trains engineering and business teams on prompt and context engineering and agent evals, and coaches applied AI engineers. Raised engineering delivery 3× at constant headcount across 7 disciplines.
1 of 4 AI ExpertsAgent HarnessesMCPToken-Cost Models
04
Education
Visitor – Physical AI
Chung-Ang University
2025 · Seoul, South Korea
Open-vocabulary 3D scene understanding, embodied AI, and the operational realities of Physical AI in hospitals, healthcare, and agriculture (KEIT government-funded).
Ph.D. in Artificial Intelligence
Seoul National University of Science and Technology
2020 – 2025 · Seoul, South Korea
Research on open-vocabulary 2D/3D instance segmentation, VLM alignment, multimodal pseudo-labeling, and person re-identification. Published in Q1 journals (Neurocomputing).
Cost-routed specialized-agent platform — token efficiency by avoiding single expensive execution paths.
AgentsCost OptimizationMulti-Agent
CareerBridge Agentic AI
Side Project
Connects Vietnamese students across 20+ South Korean universities to job recruiters — multi-agent orchestration (Gemini, PaddleOCR, Whisper, LangGraph) for community Q&A, content, and personalized career feedback.
LangGraphMulti-AgentGeminiWhisper
FastToolSearchAdapter
Open Source
Tool discovery for LLM agents; retrieves relevant tool schemas at runtime to reduce prompt bloat.
LLM AgentsTool DiscoveryRuntime
08
Open Source Contributions
MistralCoreML — Apple
Open-source PR merged into Apple's MistralCoreML pipeline, enabling optimized on-device inference of Mistral models on Apple Silicon.
CoreMLMistralAppleOpen Source
Self-Hosted AI Copilot — Ollama
Privacy-first replacement for cloud coding copilots — code never leaves the developer's machine, zero subscription cost. Self-hosted on Llama-3.1 with LangChain and Ollama.
Recognized 3× for on-time, client-quality enterprise delivery on Fortune Global 100 industrial accounts.
BoschEnterprise Delivery
10
Get in Touch
I'm open to Staff AI Engineer roles, research collaborations, and consulting on multi-agent architectures (A2A, MCP, LangGraph), harness engineering and AI-first SDLC adoption, and banking and financial AI transformation. Let's build something intelligent together.
