Hyunho Lim

@ For English

AI/ML Engineer

🇰🇷 Seoul, South Korea

Email

gusghrlrl101@gmail.com

GitHub

https://github.com/gusghrlrl101

🙋🏻‍♂️ How I Work

MLOps platform developer connecting AI/ML technology to real services. I focus on practical problem-solving by connecting technology, team, and business.
I have designed and operated end-to-end pipelines ranging from building stable and scalable machine learning platforms to AI model training-serving-monitoring.
I introduce my own way of working, refined through my experiences so far.

Quickly, Accurately, and Smartly.

I prioritize efficiency and utilize given time and resources to create the best results. I clearly set priorities and automate repetitive tasks. Also, I solve problems together through collaboration and reviews with the team.

Work Fun, Results Amazing.

I believe immersion starts from fun. I have a positive attitude of trying to find my own meaning and fun anywhere. I create a pleasant atmosphere through positive communication with team members and strive to be a colleague people want to work with.

Diverse Experiences, Different Perspectives, Innovative Results.

Diverse experiences across various technologies and domains broaden the framework for problem-solving. Through challenges, I gain new insights and can respond flexibly to change. I am not afraid of newness and always try to grow by enjoying learning.

Find, Move, and Solve.

I don't wait for problems, I find them. I deeply understand the flow of business and organization, and discover hidden problems and opportunities. I pursue sustainable improvement and value creation, not just simple solutions.

📚 Skills for Achieving Goals

I dream of becoming a versatile developer proficient in various fields, and especially aim to grow as an expert specializing in AI technology.
Below are the capabilities I am continuously developing to achieve these goals.

Main Domains: ML/AI, MLOps, Data Engineering

Tech Stack: Python, Kubernetes, MLFlow, BentoML, Airflow @View Details

Recent Interests: ML Workbench, Recommender System

👨🏻‍💻 Experience

My resume contains not just a simple career, but traces of growth through experience. I will briefly introduce the history that made me who I am today.

Woowa Brothers

AI Division, AI Platform Team

AI/ML Engineer

2022.06 ~ Present (years months)

I am developing an in-house common AI platform so that AI can be naturally applied to various services.

Design and operation of end-to-end pipelines ranging from AI model development - production environment service - monitoring.

Proactively planned and led team-level tasks, securing technical/organizational capabilities.

Contributed to productivity and quality improvement by leading collaboration culture and system structure.

I developed the ability to technically design and lead tasks, specify requirements, and convert them into a system structure.

Kakao & Kakao Enterprise

AI Lab, Vision Team

AI/ML Engineer

2019.06 ~ 2022.06 (3 years)

I was responsible for connecting the AI lab's models to actual services, focusing on inference environment configuration and optimization.

Experience in the entire process of designing–developing–deploying–monitoring small-scale AI systems.

Optimization and stabilization of AI inference environment, securing practical system operation capabilities.

Strengthened practical sense by applying various technical elements to solve real business problems.

I solidly built practical sense in connecting AI models to actual services and problem-solving capabilities.

💻 Projects

Among the projects I led, I have listed them in recent order, focusing on large-scale projects.
You can check the problem background, solution process, and results through the table below.

AI Studio - AI Platform Component Integration

I designed and developed AI Studio, a dedicated management system that integrates various scattered components within the AI platform into one, considering consistent usability and scalability.

Woowa Brothers
2024.06 ~ Present

Background	As the components constituting the AI platform were developed individually, management convenience decreased, and integrating new features incurred high costs. There were limitations in providing user accessibility and a consistent operational experience within the platform. To solve this, a central integrated system for the AI platform was needed.
Process	Based on user surveys and feedback, I created a roadmap and led the structural design considering the long-term growth direction of the platform. I personally planned and produced the entire UI/UX design using Figma, and developed a FastAPI-based backend and a React-based frontend. I configured an end-to-end environment including authentication and authorization systems, and internal platform linkage. I was able to quickly learn and apply unfamiliar environments to stably launch the entire system.
Result	By managing previously scattered components in one place through AI Studio, the consistency of platform operation and maintenance efficiency were significantly improved. Users gained faster access to features and task execution, laying the foundation for the AI platform to evolve into a product form. This system became an opportunity for future scalability with various components such as UI and CLI.

AI Platform Serving Environment Construction

Designed an automated environment for real-time AI model serving and streamlined the entire serving process from deployment to monitoring.

Woowa Brothers
2023.06 ~ 2024.06

Background	In the absence of a real-time AI model serving environment, many resources were wasted due to manual infrastructure configuration and deployment each time. As related tasks increased, service launch delays and operational burdens intensified, making the construction of a serving environment an urgent task. I proposed this task and it was promoted after discussion within the team.
Process	Investigated real-time serving demand and usability, and reviewed the tech stack within the team based on system stability, performance, and operational convenience. As a result, we adopted BentoML to build a universal and lightweight serving environment. We built an automated deployment pipeline by linking GitLab CI and ArgoCD, and secured operational stability by designing version control and rollback environments. We automated the creation of monitoring dashboards to track service status in real-time.
Result	The serving process was reduced by 40%, saving engineering resources and increasing the speed of service application. Currently, it stably processes over 2 billion requests per month and up to 5,000 requests per second for about 30 real-time services. Thanks to the platformized structure, expanding new services has also become much easier.

Detailed information is summarized on the Woowa Brothers Tech Blog (https://techblog.woowahan.com/19548).

AI Platform Training/Batch Environment Construction

Designed and built an automated environment for efficient AI model training and batch processing, including data management, execution pipelines, and environment integration.

Woowa Brothers
2022.06 ~ 2023.06

Background	At the time when there was no dedicated environment for AI training and batch jobs, much time and resources were consumed manually configuring settings and matching environments for each project. Errors frequently occurred due to environment inconsistencies, and there was a significant communication burden in the collaboration process due to differences in working methods. With model development cycles becoming shorter, building a standardized automated environment that could simultaneously improve development productivity and operational efficiency emerged as an urgent task.
Process	Organized common functions into a library and established data management policies and authorization systems. Designed a YAML-based structure for easy configuration of batch jobs. Introduced MLflow to configure experiment management and model tracking environments, and secured consistency between development and operation environments based on Docker. The entire process was based on internal team reviews and tests, and I proactively led the design and development with ownership.
Result	Reduced the training pipeline development process by about 30% and the batch job process by over 67%, significantly improving overall development and operational efficiency. Currently, more than 130 batch jobs are stably operating, laying the foundation for greatly improved productivity and collaboration efficiency.

Additional information can also be found at the Woowa Brothers Tech Seminar (https://www.youtube.com/live/MA5A7Xqb-7U).

Unmanned Convenience Store Project

Designed and developed a structure that links multiple AI models in a pipeline, enabling real-time inference in both edge device and server environments.

Kakao & Kakao Enterprise
2019.06 ~ 2022.06

Background	This was a project for building an in-house unmanned convenience store, with complex requirements to recognize human location and behavior. For this, a system capable of real-time inference by integrating various AI models such as human recognition, joint extraction, and movement tracking into a single pipeline was needed. The operating environment had to support both edge devices and server environments. As the sole developer in the team at the time, I led the design and development of the system, refining the implementation direction through internal team reviews and discussions.
Process	Designed the structure using DeepStream to integrate multiple camera inputs and pipelines, and configured it for real-time inference on edge devices through ONNX and TensorRT conversion, and CUDA optimization. Stabilized data flow between pipelines by linking Kafka, Redis, InfluxDB, etc. Repeatedly performed tuning such as model quantization, parallel processing structure, and buffering optimization for performance improvement.
Result	Implemented a system capable of real-time 3D location tracking using only general RGB cameras, and completed verification based on actual scenarios by converting an in-house test space into an unmanned convenience store. As the inference pipeline and operating infrastructure were stably linked, a foundation for operating a complex model-based AI system was established.

Detailed information can be found in the if Kakao conference video (https://tv.kakao.com/channel/3693125/cliplink/414132079).

🎖️ Activities

I believe it is important to not only proceed with work but also to actively share and exchange experiences, so I try to leave records from time to time.