SHRAG: A Framework for Combining Human-Inspired Search with RAG
Hyunseok Ryu, Wonjune Shin, Hyun Park
arXiv preprint (arXiv:2512.00772), 2025
ScienceON Challenge (RDGenAI 2025, KISTI)
[arXiv]Digital Twin Agent Implementation based on CIP Concept
Hyunseok Ryu, Inyong Song, Jong-Won Kim (류현석, 송인용, 김종원)
Annual Conference of KIPS 31(2), 463–466, 2024
Korea Information Processing Society (한국정보처리학회)
Automated Market-Trend Analysis
2022.06 – 2022.10GIST AI project · Team Lead
A semi-automated market-response system for hyper-personalization: architected the pipeline and combined multiple NLP models to extract trends from online communities.
- Architected the overall solving system
- Combined BERT, Transformer, LSTM and soynlp for the task
- Hands-on Korean-text data preprocessing
BERTTransformerLSTMsoynlpPython
Restricted-Area Human-Detection CCTV Alert App
2020.04 – 2020.06ROK Army competition support · Dev Lead & Planner
On-device object detection on a Raspberry Pi 4 detects intruders in a restricted zone, pushes an alert to a phone, and records / streams to a server in real time.
- React Native app with push notifications, GPS, and Kakao API
- ExpressJS / Node.js RESTful and socket server
- TensorFlow object detection running on Raspberry Pi 4
- Deeper understanding of JavaScript and of individual ownership within a team
JavaScriptReact NativeNode.js/ExpressTensorFlowRaspberry Pi
Fatogo — Group Travel Booking Platform
2018.04 – 2018.12Funded by GIST (₩8M) · Dev Lead & Service Planner
A tailored accommodation-booking platform for groups of young travelers.
- Deployed a Django backend on AWS EC2
- Designed RESTful APIs and integrated PostgreSQL with Django
- Built and led a 4–6 person team
- Learned first-hand how much project planning and communication matter
PythonDjangoPostgreSQLAWS EC2HTML/CSS
Self-Balancing Robot
2018.07 – 2018.08GIST Competition · Individual project
Kept a robot upright from MPU 6-axis sensor data. Because gimbal lock does not occur in this setup, a quaternion frame was unnecessary.
- Complementary filter over the MPU 6-axis sensor for accurate attitude estimation
- PID control for balancing; raised the center of mass for stability
- Motors were too slow to fully prevent falls — tuning the PID integral (I) gain remained the open challenge
mbed STM32F411CPID control
Vehicle Tracking System
2018.07 – 2018.08GIST Competition · Team Lead / Developer
Collected and analyzed gyroscope and accelerometer data from an Android background process to infer which mode of transportation a user was riding, with an analysis pipeline on AWS.
- Android background data-collection app (Android Studio)
- Django REST API and data-collection server (AWS EC2, SQLite)
- Real-time analysis with Python multiprocessing
- First real experience of team collaboration and project planning
AndroidDjangoPythonAWS EC2Ubuntu