SEHUN KIM

Project 1: Integrated ML Inference Platform

Main Responsibilities: Platform development and operation
Role: Overall platform development lead, platform development, ML module development
Technologies Used: Python, SQLAlchemy, PyTorch, TensorFlow, MTCNN, PyAudio, NumPy, MySQL, SQS, K8s
Project Duration: February 2020 – November 2023 (1 year 9 months)
Team Size: 2 members
Project Details:

• Objective: Integrate existing ML inference engines from various in-house services (AI recruitment service, interactive AI content service, mental health service)
  Minimize operational costs and reduce inference latency
• Achievements: Transitioned ML inference modules to be managed via the database, allowing for rapid construction of new ML Inference pipelines using the module combinations during new service launches. Previously, new service required over 3 months for ML inference pipeline preparation; now completed within 2 weeks.
  Reduced overall ML inference pipeline operating costs by more than 2/3 after migration to the integrated inference platform.
  Reduced inference latency from 5 minutes to under 3 minutes using parallel pipelines.
  Applied auto-scaling to reduce operational costs by 5% or less during spike traffic events.
  

Project 2: Agent Builder

Main Responsibilities: Backend development
Role: Backend DB design and interface, asynchronous agent worker development
Technologies Used: Python, FastAPI, BentoML, VLLM, Langchain, Strawberry, SQLAlchemy, MySQL, PostgreSQL, RabbitMQ, ScyllaDB, Milvus, ECS, Terraform, K8S, GitHub Actions
Project Duration: November 2023 – Present
Team Size: 4+ members
Project Details:

• Objective: Develop an agent creation platform
  Build a no-code / low-code agent development environment
  Provide agent testing features
• Achievements: Completed workflow creation feature similar to n8n and Make
  Developed API integration functionality for generated workflows
  Completed workflow testing and debugging features
  

Project 3: MLOps Implementation

Main Responsibilities: Data preprocessing pipeline development and MLFlow-based centralized training platform
Role: Developed data preprocessing pipeline using K8S + Airflow, centralized model training management using MLFlow + TorchLightning
Technologies Used: Python, K8S, Airflow, NumPy, Pandas, MLFlow, TorchLightning
Project Duration: January 2023 – February 2023
Team Size: 1 member
Project Details:

• Objective: Improve script execution method for each stage of the preprocessing pipeline
  Enhance individual researcher learning management methods
• Achievements: Built a preprocessing pipeline using Airflow DAGs, enabling simple execution and monitoring via a dashboard
  Centralized logging of training processes with MLFlow, managing research code, settings, and artifacts. => Eliminated issues of experiment result loss caused by personnel changes or departures
  

Project 4: Data Management System Development

Main Responsibilities: Development of learning data management system, labeling system, and inference result monitoring system
Role: Backend and frontend development
Technologies Used: SpringBoot, ReactJS
Project Duration: February 2020 – May 2023
Team Size: 4 members
Project Details:

• Objective: Build a systematic management system to ensure AI reliability
  Develop a flywheel for AI learning data
  
• Achievements: Developed various data evaluation tools to ensure data reliability and quality assurance
  Implemented data versioning for training purposes, allowing for historical tracking of training data
  Controlled data access to prevent data contamination and ensure privacy protection => Achieved TTA AI Reliability Certification in 2023
  

Project 1: E-commerce Web Service Development and Operation

Main Responsibilities: Overall web service development
Role: Backend, frontend, and admin development
Technologies Used: Python, Django, Vue.js, MariaDB, Celery, MongoDB
Project Duration: August 2016 – January 2020 (3 years 5 months)
Team Size: 1 member + (up to 2 members)
Project Details:

• Objective: Ensure website stability
  Increase sales through data-driven marketing
  Improve service operation efficiency, focusing on product and shipment management features
• Achievements: Turned around from over 50% deficit to break-even within 6 months and achieved more than double the break-even sales within 2 years
  Eliminated service disruptions, reducing occurrences from 2-3 times a week to zero downtime
  Reduced webpage loading time from over 3 seconds to under 300ms
  

Project 2: Remarketing Automation Platform Development

Main Responsibilities: Event collection and remarketing automation backend, Spark query development
Role: Overall development lead, backend, Spark query development
Technologies Used: Python, Flask, MariaDB, MongoDB, Spark
Project Duration: August 2019 – December 2020 (5 months)
Team Size: 1 member + (up to 1 member)
Project Details:

• Objective: Track all events both on client and server sides
  Perform customer funnel analysis
  Create target segments and perform remarketing based on collected logs for performance marketing
• Achievements: Developed a web front SDK for customer tracking
  Developed a funnel analysis feature using Amplitude
  Created a function to extract target segments based on RFM (Recency, Frequency, Monetary) metrics
  

Project 1: WouldU Coffee

Main Responsibilities: Service development
Role: Backend, Android, iOS, and admin web development
Technologies Used: JavaScript, Node.js, Express, Kotlin, Swift, MariaDB
Project Duration: October 2015 – July 2016 (9 months)
Team Size: 1 member
Project Details:

• Objective: Develop a location-based indie cafe ordering app service
  Develop user apps for Android and iOS
  Develop a web app for cafe management
  Develop a web app for service management
• Achievements: Developed a web frontend framework
  Implemented mobile app animations
  Successfully launched the service
  

Project 1: ShapeEngine

Main Responsibilities: 3D printer sharing web service development
Role: Backend and frontend development
Technologies Used: Python, Django, MariaDB, JavaScript
Project Duration: May 2015 – September 2015 (4 months)
Team Size: 5 members
Project Details:

• Objective: Develop a 3D printer sharing web service
  Enable 3D printing from user-uploaded models
  Implement model slicing and direct transmission to 3D
• Achievements: Developed a model viewer using Three.js
  Built a custom web framework
  Successfully launched the service
  

Project 1: Cloud Storage Development

Main Responsibilities: Storage backend development and infrastructure management
Role: Spring server development and GlusterFS distributed storage operation
Technologies Used: Java, Spring, MyBatis, MySQL, GlusterFS
Project Duration: March 2013 – March 2014 (12 months)
Team Size: 6 members
Project Details:

• Objective: Develop a Dropbox-like service
• Achievements: Built distributed storage using GlusterFS
  Developed a file management system using Spring
  

Project 2: Whoopee

Main Responsibilities: Serverless file sharing app development
Role: Android app development
Technologies Used: Java, Android, mDNS
Project Duration: April 2014 – April 2015 (12 months)
Team Size: 5 members
Project Details:

• Objective: Develop local network file sharing functionality
  Implement remote file sharing functionality
• Achievements: Developed a local network file transfer app using mDNS
  Implemented proof of concept (PoC) for remote file sharing using hole punching
  

Project 3: Android/HTML5 Set-Top Box Development

Main Responsibilities: app launcher for Set-top box development
Role: Android set-top box development, HTML5-based set-top box development
Technologies Used: Java, Android, HTML5, RequireJS
Project Duration: March 2013 – April 2015 (13 months)
Team Size: 4 members
Project Details:

• Objective: Develop an Android launcher for set-top boxes
  Build an HTML5-based set-top box
• Achievements: Developed an Android-based launcher for set-top boxes, including animation and remote control interaction features
  Developed a HTML5-based set-top box using RequireJS
  

Project 1: Cloudit v1 Development

Main Responsibilities: Virtualization platform development
Role: API Gateway, authentication, and monitoring development
Technologies Used: Java, Spring, iBatis, ActiveMQ, MySQL, Xen Server, JavaScript, jQuery
Project Duration: July 2010 – December 2011 (17 months)
Team Size: 6 members
Project Details:

• Objective:Develop a service similar to AWS
  Build a web console for the service
• Achievements:Developed an IaaS platform based on Xen Server (first in Korea)
  Launched the Cloudit service
  Developed an interactive web console using jQuery
  

Project 2: Cloudit v2 Development

Main Responsibilities: Virtualization platform package development
Role: Virtualization engine, distributed storage, monitoring agent development, Linux packaging
Technologies Used: C++, Boost, Thrift, Xen Hypervisor, Open vSwitch, GlusterFS, Pacemaker
Project Duration: January 2011 – August 2012 (17 months)
Team Size: 3 members
Project Details:

• Objective: Redesign the engine based on Xen Hypervisor
  Reimplement using C++
  Package the solution in the form of a Linux distribution
• Achievements: Developed volume management based on VHD
  Managed VLANs using Open vSwitch, isolating virtual networks per user
  Developed image management storage using GlusterFS
  Developed a custom virtualization engine and collected metrics through a monitoring agent
  Optimized and compiled the Xen Hypervisor from linux kernel and packaged it
  

Project: Security Threat Detection, Analysis, and Response for KB Bank

Main Responsibilities: On-site support at KB Bank
Role: Malware analysis and response for bank branch and internal staff, security threat analysis and response
Technologies Used: C, Python, Immunity Debugger
Project Duration: May 2018 – January 2020 (1 year 8 months)
Team Size: 6 members
Project Details:

• Objective: Minimize damage caused by malware
  Prevent and respond to bank system breaches
• Achievements: Reported the highest number of new malware cases within the team
  Successfully restored systems within 20 minutes during the nation-wide DDoS attack on July 7, 2009, while other financial institutions faced outages for over a day, and some public institutions were down for up to a week