Interactive

Jaehyeok Choi

AI Engineer

Seoul, South Korea wowogur12@naver.com github.com/choijhyeok linkedin.com/in/choijh0119

Profile

Forward-deployed AI Engineer who works directly in financial, insurance, and manufacturing environments to turn business problems into production AI systems. Specializes in RAG pipelines, Agent orchestration, durable workflows, and realtime agents, from PoC validation to production deployment.

Work Experience

AI Engineer | KT Seoul, South Korea
Jul 2024 — Present
  • Embedded in enterprise client projects to translate business workflows and constraints into production AI systems
  • Delivered production RAG services and designed Multi-Agent systems across insurance, finance, and manufacturing
  • Owned RAG indexing, LangGraph orchestration, realtime Function Calling, and Azure/AKS integration boundaries
Data Scientist | AIFactory Seoul, South Korea
Apr 2022 — May 2024
  • Designed and validated AI competitions and enterprise PoCs across LLM, CV, and time-series domains
  • Built LLM fine-tuning pipelines, RAG systems, and performance evaluation workflows (RAGAS)
  • Implemented large-scale data preprocessing and automated inference pipelines
Data Scientist | GSITM Seoul, South Korea
Jul 2021 — Jan 2022
  • Developed forecasting and optimization models for retail and logistics domains
  • Implemented demand forecasting and matching algorithms for real-world business operations

Key Projects (KT)

Master Agent Engineer, AI Virtual Assistant (Hanwha General Insurance)

May 2026 — Dec 2026
  • Designing the LangGraph-based Master Agent for an AI assistant serving approximately 20,000 insurance planners
  • Own Graph, State, intent routing, execution sequencing, context handling, and fallback policies
  • Orchestrate product recommendation, policy search, underwriting guidance, UI control, and core-system tools
  • Apply PII detection, input validation, timeout, retry, duplicate-call prevention, and normalized error responses
  • Integrate with Azure Landing Zone, AKS, ExpressRoute, and on-premise GitLab delivery environments

AI Engineer, Large-scale Insurance Policy RAG Pipeline (Meritz)

May 2025 — Dec 2025
  • Converted approximately 390,000 third-party insurance policy documents into searchable assets
  • Built an Azure Durable Functions indexing workflow with fan-out/fan-in, retry, throttling, and state tracking
  • Compared six chunking strategies against insurance-domain questions and selected Parent–Child RAG
  • Preserved PDF page and table structures through PyMuPDF-based parsing and chunking
  • Deployed to production; used by approximately 1,000 insurance planners daily

AI Engineer, AI Branch – Realtime Voice-based AI Banker (Shinhan Bank)

Jan 2025 — Apr 2025
  • Built a voice-to-voice AI banker PoC using the GPT-4o-Realtime API
  • Implemented WebSocket-based real-time streaming pipelines
  • Developed loan recommendation agents using Function Calling
  • Integrated Azure AI Search for financial product retrieval and reasoning
  • Deployed the full system on Azure Container Apps

AI Engineer, Investment Agent PoC (POSCO)

PoC
  • Normalized PPTX, DOCX, XLSX, and PDF investment documents into a unified PDF processing path
  • Combined Azure Document Intelligence OCR and Vision Models to enrich tables, images, and charts
  • Applied PageIndex and Keyword + Vector Hybrid Search to preserve document and slide hierarchy
  • Built workflows to retrieve similar investments, review issues, Q&A, and decision rationale
  • Validated retrieval at document and page level without overstating unconfirmed quantitative metrics

AI Engineer, Web Search Agent (Giga Genie / JTS Thailand)

Sep 2024 — Mar 2025
  • Developed multiple Web Search Agent PoCs based on Bing Search
  • Designed agent workflows using LangGraph
  • Reduced hallucinations using Self-RAG and Corrective-RAG
  • Achieved up to 94% answer accuracy with sub-10s response latency
  • Built demo applications using Gradio

AI Engineer, Legal RAG System PoC (Korea Forest Service)

Aug 2024 — Dec 2024
  • Built a legal-domain RAG system PoC for statutes and legal precedents
  • Designed a combined LLM fine-tuning + RAG architecture
  • Achieved Top-5 retrieval accuracy of 93.51%
  • Evaluated performance with RAGAS (Context Precision, Recall, Faithfulness > 90)
  • Developed a demo web using Gradio and pdf.js with source highlighting

Selected Projects (AIFactory)

Data Scientist, LLM Fine-tuning and RAG Evaluation

  • Fine-tuned LLaMA2, Gemma, and EEVE models using QLoRA and DeepSpeed
  • Built RAG pipelines and evaluated performance using RAGAS
  • Developed crawling pipelines for external data collection

Data Scientist, AI Competition Design (CV / Time-Series)

  • Designed challenges for Object Detection, Segmentation, and Pose Estimation
  • Built datasets in COCO format and defined evaluation metrics (Macro F1, IoU, MAE)
  • Led competition design and validation for multiple enterprise and public clients

Selected Projects (GSITM)

Data Scientist, Retail Sales Forecasting System

  • Preprocessed convenience store sales data with seasonality analysis
  • Built Prophet-based sales forecasting models
  • Proposed inventory optimization strategies based on forecast results

Data Scientist, Genetic Algorithm-based Logistics Matching

  • Implemented optimization algorithms using DEAP
  • Developed matching logic to maximize profit under weight and cost constraints
  • Improved logistics efficiency and operational decision-making

Personal Projects & Talks

Open Source Maintainer, Korean HWP/HWPX Document Parsing Open Source

  • Developed Python libraries for parsing HWP and HWPX documents
  • Addressed real-world limitations of Korean document processing in Python environments
  • Resources: GitHub Repo, LinkedIn Post

Personal Project, Gemma Function Calling Assistant

  • Fine-tuned Gemma 7B using SFT to enable Function Calling
  • Built a personal assistant that executes external tools and delivers results via KakaoTalk

Speaker / Open Source Contributor, LLaMA2 Fine-tuning & RAG Pipeline

  • Presented an end-to-end LLaMA2 fine-tuning and RAG pipeline
  • Implemented QLoRA-based instruction fine-tuning optimized for Colab T4
  • Built an easy-to-use fine-tuning pipeline with a Gradio UI
  • Resources: GitHub Repo, YouTube Talk

Speaker, Building RAG-based Services with Streamlit & LangChain

  • Speaker at LangChain KR Meetup (2024 Q1)
  • Demonstrated deployment of RAG-based AI services using Streamlit
  • Explained vector DB separation and QA-chain-based recommendation architectures
  • Showcased automated report generation pipelines (HTML → PDF)

Education

Gyeongsang National University

B.S. in Information Statistics and Computer Science

Mar 2016 — Feb 2022

South Korea

Skills

Languages
Python JavaScript SQL
LLM / GenAI
RAG LangChain LangGraph ReAct Fine-tuning Function Calling PageIndex Hybrid Search RAGAS
Infrastructure
Azure Functions Azure AI Search Azure Container Apps AKS Azure Landing Zone Docker
Document Processing
PyMuPDF OCR Azure Document Intelligence Vision Model Large-scale Document Pipelines