Skip to content
View leejulie05's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report leejulie05

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
leejulie05/README.md

이승연 (Seungyeon Lee)

Sogang University, Department of Artificial Intelligence (3rd year) NLP · LLM Systems 신뢰성 있고 안전한 AI 시스템 설계에 관심이 있습니다.

📫 jjulie2005@gmail.com


관심 영역 (Areas of Interest)

  • LLM 신뢰성 / Reliability — Hallucination 방지, grounding, 출력 검증
  • AI Safety — Operational guardrails, content safety, ethical design
  • Domain-specific NLP — 한국어 NLP, 도메인 지식 + 메트릭 설계

프로젝트마다 hallucination 방지·data privacy·guardrails 같은 요소를 시스템 설계 단계부터 함께 고민합니다.


대표 프로젝트 (Featured Projects)

Multimodal AI Personal Trainer — Custom LSTM (자세 분류) + GPT-4o (피드백 챗봇). 의료·프라이버시·운영 측면 Ethics & Safety 가드레일을 시스템 설계 단계부터 포함.

PyTorchLSTMMediaPipeGPT-4oStreamlitPrompt Engineering

정부출연연구기관(STEPI) 의뢰로 단독 구현한 채용 자기소개서 분석 NLP 시스템. 온프레미스 환경에서 Qwen3-14B + RAG로 hallucination 방지·grounding 설계. (코드 비공개, 방법론 정리)

Qwen3-14Bjina-embeddings-v3RAGKorean NLPOn-Premise LLM

DeepSeek-R1 교사로 KorMedMCQA train 에 long-CoT(reasoning) trace 를 생성하고, **반사실 추론 충실성 필터(Inverse-MATCHA)**로 "답을 바꿔도 똑같이 그럴듯하게 정당화하는 사후합리화 CoT"를 걸러낸 뒤 Qwen3-8B student(LoRA)를 학습·평가


Tech Stack

Languages: Python · C ML / NLP: PyTorch · Transformers · LangChain · sentence-transformers · scikit-learn LLM: Prompt Engineering · RAG · fine-tuning · LoRA Data: pandas · numpy · matplotlib · MediaPipe Tools: Streamlit · Jupyter · Git


About me (English)

3rd-year undergraduate at Sogang University, Department of Artificial Intelligence.

Interested in NLP, LLM systems, and AI Safety, with a focus on building reliable AI systems — incorporating hallucination prevention, data privacy, and operational guardrails into system design from the start.

Recent work spans on-premise NLP pipelines for government recruitment evaluation, multimodal AI coaching with embedded safety guardrails, and custom-metric design for cultural data mining.

📫 jjulie2005@gmail.com

Pinned Loading

  1. SquatPT-LLM-ChatbotSquatPT-LLM-ChatbotPublic

    SquatPT: Multimodal AI Personal Trainer. Real-time squat analysis using Custom LSTM & GPT-4o. (Features: Temporal Interpolation, Smoothing, LLM Integration)

    Python

  2. recruitment-eval-systemrecruitment-eval-systemPublic

    NLP pipeline for resume screening & interview question generation (Qwen3-14B + RAG, on-premise). Methodology write-up only — code private.

  3. Gugak-Modern-Music-data-mining-projectGugak-Modern-Music-data-mining-projectPublic

    한국 전통 음악(국악)의 정서적 특성을 정량 점수로 정의하고, 멜론 차트 26,000곡과 비교해 어느 가수·곡이 국악과 가장 닮았는가를 데이터 마이닝으로 찾는 프로젝트.

    Jupyter Notebook

  4. AITEAM444/KoMEDMCQA_CoTAITEAM444/KoMEDMCQA_CoTPublic

    Python