JAY

"Exploring human essence and creating Übermensch through AI."

Personal Information

Jeong Jae Heon

🧑🏻‍💻 Information

Birth: 1999. 01. 19

E-mail: heonyus@gmail.com

GitHub: heonyus

🪖 Military Service

Period: 2020.04 - 2021.10 Auxiliary Policeman (Seoul Metropolitan Police Agency, 3rd Mobile Unit)

📚 Education

2024.08 MJU, Graduated Summa Cum Laude in Industrial Management Engineering

2022.03 MJU, Transferred to Industrial Management Engineering

2019.02 DIMA, Withdrew from Acting Major

2018.03 DIMA, Entered Acting Major

2017.02 Gwan-ak High School, Graduated

2014.03 Gwan-ak High School, Entered

Stacks

Projects

Development of Personal Knowledge Model to Understand User's Persona

Developed a model that extracts and manages personal knowledge about learners, allowing the AI to engage in more personalized conversations during tutoring sessions.

Period: Sep. 2023 - Present

Key Achievements:

Implemented a knowledge extraction module with GPT-3.5 API.
Optimized personal knowledge management using a multi-domain approach.
Improved AI tutoring personalization through effective knowledge integration.

Technologies:

GPT-3.5Personal Knowledge ManagementPythonMLOps

Open-end Chat Modifier Model Development and Deployment

Built and deployed a model that modifies the output of AI chat systems, enabling the generation of more context-aware follow-up questions in conversations.

Period: Jan. 2024 - Mar. 2024

Key Achievements:

Achieved efficient context handling in long AI conversations.
Deployed the model using AWS SageMaker with a rapid response time of 1.7 seconds.
Implemented a CI/CD pipeline for continuous model updates.

Technologies:

GPT-3.5AWS SageMakerPythonDocker

Development of National Assembly Bill Proposal System Using Existing Legislation

Developed a system for the South Korean National Assembly that assists in proposing bills by utilizing existing legislation, contributing to the legislative process.

Period: Oct. 2023 - Dec. 2023

Key Achievements:

Automated the bill proposal process using NLP techniques.
Enhanced legislative efficiency by integrating existing laws into the system.
Successfully deployed and tested the system within the National Assembly.

Technologies:

NLPPythonAPI Development

Entity Recognition Search Model Development and LLMOps Engineering System

Engineered a system that evaluates and recommends words based on learner input by embedding their utterances and comparing them with a pre-built word database using cosine similarity.

Period: Sep. 2023 - Dec. 2023

Key Achievements:

Developed a robust entity recognition model with SBERT.
Improved word recommendation accuracy by over 75%.
Integrated LLMOps for automated deployment and scaling.

Technologies:

SBERTGPT-4 APILLMOpsDockerAWS

Development of NER - Search Model to Assess Learner's Familiarity with Words and Expressions

Created a Named Entity Recognition (NER) search model to evaluate and improve learners' familiarity with words and expressions, integrating the system into educational tools.

Period: Sep. 2023 - Dec. 2023

Key Achievements:

Successfully assessed and improved learner vocabulary.
Incorporated advanced NLP techniques to enhance learning outcomes.
Streamlined integration with existing educational tools.

Technologies:

NERSBERTPythonLLMOps

Development of Dynamic Prompting that Changes System Messages Based on Textbook Topics

Implemented a dynamic prompting system for educational purposes that adjusts system messages based on textbook topics, improving user interaction with the AI.

Period: Sep. 2023 - Nov. 2023

Key Achievements:

Enhanced AI interaction by dynamically adjusting system prompts.
Reduced user confusion by aligning prompts with textbook content.
Improved user satisfaction in educational AI applications.

Technologies:

GPT-3.5Prompt EngineeringPython

Development of DX-ASTI Demand-based Service Model Recommendation System

Developed a recommendation system for ASTI (KISTI) that uses pre-trained models like LLaMa2 and various key technologies (KeyBERT, SBERT, CML, and contrastive learning) to extract tailored keywords and enhance service matching accuracy for member companies based on their profiles.

Period: Mar. 2023 - Aug. 2023

Key Achievements:

Successfully fine-tuned LLaMa2 for ASTI member company profiles.
Improved service matching accuracy through tailored keyword extraction.
Implemented a robust contrastive learning model for enhanced recommendation.

Technologies:

LLaMa2KeyBERTSBERTCMLContrastive LearningPython

Development of Heterogeneous Domain Integrated Graph Neural Network for Topic Tomography

Created a RAG (Retrieval-Augmented Generation) system for KISTI, which identifies topics within documents and finds the keywords that constitute those topics by analyzing similarities across documents.

Period: Jan. 2023 - Aug. 2023

Key Achievements:

Developed a hybrid embedding structure for improved topic identification.
Implemented a contrastive learning approach for accurate topic clustering.
Enhanced document similarity detection through advanced graph neural networks.

Technologies:

SentenceBERTClusteringContrastive LearningPython

Travel Itinerary Scheduling Using Reinforcement Learning and NLP

Designed and implemented a scheduling system for travel itineraries by using reinforcement learning and natural language processing techniques.

Period: Mar. 2023 - Jun. 2023

Key Achievements:

Developed an efficient scheduling algorithm using reinforcement learning.
Integrated NLP techniques for context-aware itinerary adjustments.
Successfully tested and validated the system with real-world data.

Technologies:

Reinforcement LearningNLPPython

Development of Reinforcement Learning Model for Power Generation Fuel Consumption Optimization

Developed a reinforcement learning model to measure fuel consumption characteristics and optimize combustion processes, contributing to more efficient power generation.

Period: Oct. 2022 - Dec. 2022

Key Achievements:

Optimized fuel consumption in power generation through reinforcement learning.
Implemented a model that significantly reduced operational costs.
Achieved measurable improvements in combustion efficiency.

Technologies:

Reinforcement LearningPython