AI Engineer who specializes in Language model and Web technology.
Projects
RTSum
2023.3 ~ 2023.8
- Submitted a summarization method and working demo as a demonstration paper to EMNLP 2023.
- Minimized the hallucination problem by decomposing sentences into smaller units and recomposing them with the most important information.
- Integrated the concept of Knowledge Graph and Relation triple into summarization AI.
- Combined extractive summarization and abstractive summarization.
LLaMa2 GPTQ
2023.5 ~ 2023.6
- Chat AI which can provide responses with reference documents by Prompt engineering over vector database.
- Attained local AI without requesting external API by optimizing inference performance with GPTQ 4bit model quantization.
- Suggestion of related web pages is provided through the integration with my previous product, Texonom.
Texonom
2020.6 ~ 2023.4
- Reduced data fetching time 90% than official API method with hacking Notion.so API data structure.
- Published 10,000-page web serving knowledge graph system by manipulating Notion.so API.
- Saved 90% on server costs by migrating deployment environment from Okteto Kubernetes to Vercel.
Pointland
2020.3 ~ 2020.10
- Developed a touchscreen joystick that allows users to wander anywhere in 3D space.
- Accomplished full screen web experience with pointcloud data served from Google Cloud Storage using PWA.
Intuiter
2019.10 ~ 2021.11
- Solved the complexity of existing shortcut applications (Vim, Emacs) with an easier usage.
- Increased universality to work in any software by letting the app run in the background.
- Bound Electron and AutoHotKey by including AHK compiler in the installer which make AHK script run dynamically.
- Added a smooth mouse control on keyboard by calling Windows mouse API DLL.
Work Experience
Kakao Mobility
Software Engineer (2021.12 ~ 2022.9)
- Led a project developing of a web application project to draw 3D vector map for Autonomous driving.
- Improved vector drawing speed to extract normal vectors of a surface using 3D point cloud data from LiDAR.
- Integrated development experience of Mac, Windows, and GKE using Kubernetes on Rancher Desktop with a localized production stack (Redis, Node.js, InfluxDB, and Grafana).
- Reduced build time by 70%, simplified dependency management, and merged Node.js-based repositories into a mono-repo.
- Enhanced maintainability with unit test (Vitest) and code coverage including frontend source code by mockup browser rendering.
- Migrated Docker based application to GKE with a Google Cloud DevOps Award winning team.
- Achieved real-time error notification (Slack) and task monitoring (Grafana Map dashboard) via logs in InfluxDB.
- Refactored C++ camera projection library to Rust which convert 3d point to a 2d point using lens information.
Stryx
Software Engineer (2019.11 ~ 2021.12)
- Obtained LiDAR sensor control of vehicles with Socket.io and real-time pointcloud rendering via three.js.
- Cut bandwidth and response time by 80% with multi-level caches (Redis, Nginx Cache, and Browser Cache).
- Downsized Docker image from 1GB to 200MB using multi-stage builds which accelerated the deployment process.
- Arranged system architecture (Geoserver, PostGiS, Potree) for rendering thousands of 2D/3D vector data inside web app.
- Raised precision of drone coordinate positioning with a tuning algorithm that uses the direction of landmarks in panorama photos.
- Upgraded service reliability by migrating bare metal servers to OpenStack through building docker containers for apps.