Hi,
I'm Hyston
A
Full-Stack AI Engineer
Download CV
Professional Summary
Full-Stack AI Engineer with three years of academic research experience and over five months of industry experience building discriminative models, generative models, and LLM-based systems, including recommendation systems and Retrieval-Augmented Generation (RAG). Experienced in full-stack development using React, Next.js, and the MERN stack. Focused on deploying AI solutions in industrial environments to solve real-world problems and drive business growth.
Skills
Profesional Skills
Solid foundation in machine learning, deep learning, and data analysis, spanning both theoretical concepts and hands-on implementation.
Projects
MI-NOW — Enterprise Business Intelligence Platform
Deployed an enterprise Business Intelligence platform delivering ETL and RAG APIs for Korean manufacturing and global industrial data. Implemented a Hybrid RAG architecture combining semantic search, keyword matching, and file-based retrieval. Achieved 97.7% precision, 70.0% recall, 82.3% F1 score, with an average of 2.3% hallucination rate. Architected a dual-provider AI inference system with on-premise Ollama and CLOVA Studio fallback.
Tools: Node.js, TypeScript, Express.js, Qdrant, Ollama, CLOVA Studio, AWS (MySQL RDS, S3, OpenSearch), Redis, BullMQ, Next.js, React, Tesseract.js
SRT Voice of Customer (VoC) — AI Customer Service Platform
Contributed to developing an AI-powered VoC platform for Korea's SRT (Super Rapid Train), automatically classifying and responding to customer complaints across multiple departments using Tree-RAG (T-RAG). Built the T-RAG pipeline with entity detection, Milvus vector search, LLM reranking, and response generation, achieving 95% precision in retrieval and response selection. Implemented a multi-node LLM proxy with GPU/CPU load balancing.
Tools: Python, FastAPI, Milvus, LangChain, Tree-RAG, GPU/CPU Load Balancing
AI Resume Assistant (Personal Portfolio Chatbot)
Built and deployed an AI Resume Assistant using Retrieval-Augmented Generation (RAG), enabling recruiters to interactively query my CV, expertise, and experience. The chatbot provides personalized responses leveraging RAG techniques to retrieve relevant information from my documents and GitHub repositories. This project showcases skills in AI-driven conversational agents and enhances user engagement.
Tools: Next.js, TypeScript, LangChain, OpenAI, Qdrant, Firebase, Vercel.
XR Twin-based Rehabilitation Training Content Technology
Contributed to this project (IITP/MSIT-funded, Project No. 2022-0-00218), focusing on AI-driven rehabilitation technologies. Performed data preprocessing, analysis and visualization. Developed a hybrid model for heart rate prediction combining Dynamic Bayesian Networks (DBNs) and LSTMs. Achieved an average of 5.1 BPM MAE in prediction accuracy. This model assists in XR Twin project in AI custom coaching by providing personalized recommendations based on heart rate data.
Tools: Python, PyTorch, Jupyter, NumPy & Pandas, Matplotlib & Seaborn, scikit-learn.
Feature Selection Tool
Developed a Python-based Feature Selector Tool for automated data preprocessing, feature importance analysis, and visualization, supporting classification and regression tasks. This tool streamlines the feature selection process, enhancing model performance and interpretability. It includes functionalities for data loading, preprocessing, feature importance ranking, and visualization.
Tools: Python, Scikit-Learn, Feature-Engine, Seaborn, Matplotlib.
Work Gallery
Education
MSc in Computer Science and Engineering
Soongsil University, Seoul, South Korea (2022 – 2025)
Specialized in Machine Learning and Deep Learning and Research.
Thesis: "A Multi-Model Machine Learning Framework for Personalized Fitness Recommendations
Using DBNs and LSTMs"
BSc in Information Communication Technology
Daeyang University, Lilongwe, Malawi (2017 – 2021)
Focused on systems design and development.
Final Year Project: Designed and implemented an AI-powered tool to translate speech and text into sign language, enhancing communication for deaf students.
Publications
A Hybrid Approach to Modeling Heart Rate Response for Personalized Fitness Recommendations
📘 Journal: Electronics, Vol. 13, Issue 19 (2024)
Abstract: This paper proposes a hybrid model combining DBNs and LSTMs to personalize heart rate prediction for adaptive fitness guidance.
ProAdaFs: Probabilistic and Adaptive Feature Selection in Deep Recommendation Systems
📍 Conference: ICOIN 2024, Vietnam
Abstract: ProAdaFs introduces a probabilistic adaptive feature selection layer within deep recommendation models to improve both performance and interpretability.
Deep Adaptive Feature Selection in Deep Recommender Systems
📍 Conference: Korean Society of Information Science, Jeju (2023)
Abstract: This work explores a deep learning architecture embedded with adaptive feature selection for scalable and efficient recommendation systems.
Contact
Message Sent ✅
Thanks for reaching out! I'll get back to you shortly.