NGUYEN
QUANG DUNG
AI engineer building systems that run in the real world — not just in notebooks. Currently at Koidra AI, where physics constraints meet real-time greenhouse control.
About
Most AI projects die between the notebook and production. I work in that gap.
At Koidra, I build physics-informed models that control real greenhouses — where a wrong prediction means wasted water or stressed crops. At A-Star, I architected multi-agent platforms that needed to handle real money and real latency constraints. At BKAI, I research agentic memory architectures because LLMs that forget everything between turns are not useful agents.
The common thread: models that need to be reliable, fast, and grounded in domain constraints. Not demos. Systems.
Professional Experience
Koidra AI
AI Engineer · Remote
- Designed a hybrid control system that combines PID controllers with gradient-based anomaly detection for greenhouse climate regulation — processing sensor data in real-time to automatically adjust ventilation, irrigation, and heating.
- Automated temperature/humidity regulation across production greenhouses, replacing manual monitoring with continuous sensor-driven control and adaptive thresholds.:
- Refactored the core control module with State Machine patterns — eliminating race conditions between control loops and making the system easier to maintain and extend.
- Developed a Physics-Informed ML model for transpiration dynamics that encodes thermodynamic constraints directly into the loss function — producing physically consistent predictions that pure data-driven models could not achieve.
A-Star Group
AI Engineer (Promoted from Intern) · Ha Noi, Vietnam
- Led technical R&D for Web3 AI initiatives — translating ambiguous business requirements into production architecture, managing the full cycle from feasibility research to deployment.:
- Architected a Web3 Multi-Agent Platform using MCP (Model Context Protocol) — building the core engine for automated yield optimization that became the flagship product's primary differentiator.
- Built a crypto-wallet classification microservice with FastAPI — handling the full ML lifecycle from data curation and labeling to model training and production inference.
- Implemented RAG and NER pipelines for blockchain data — grounding LLM responses in verified on-chain data to reduce hallucination on domain-specific queries.
NLP Lab — BKAI (HUST)
Undergraduate Research Assistant · Hanoi, Vietnam
- Designed Agentic Memory architectures for LLM agents — implementing persistent context storage and retrieval mechanisms that enabled multi-turn reasoning across complex task flows without prompt bloat.
- Developed LLM distillation pipelines that compress large teacher models into compact students — reducing inference cost while retaining task performance on domain benchmarks.
- Built a reasoning evaluation framework for analyzing chain-of-thought strategies — identifying systematic failure modes in CoT prompting that inform better prompt engineering practices.
Technical Expertise
Languages
AI/ML Domains
Frameworks & Libraries
MLOps & Engineering
Honors & Awards
Silver Medal
National Physics Olympiad (VPhO)
Top 5 Finalist
DataFlow 2025 — National Data Analysis Hackathon
Education
Hanoi University of Science and Technology (HUST)
B.Sc. in Computer Science
2023 — Present