Search
๐Ÿ‘จ๐Ÿปโ€๐Ÿ’ป

About

Daniel Kim

๊น€์„ฑํ™˜ Sunghwan Kim ML Engineer at Scatter Lab. sunghwan519@hotmail.com
Working as ML Software Engineer at Scatterlab (South Korea). Interested in ML Serving Infra, LLM Serving Optimization, MLOps, and Data Pipelines.
also interested in... Music and Audio Processing
Resume
70.3KB

Career

Scatter Lab | ML Software Engineer

2021.08 ~
Handled all ML engineering tasks, including model optimization, ML serving infrastructure, MLOps pipeline, data engineering, and internal tools development for LLM research, training, deployment, and operations.
Optimized LLM serving infrastructure for AI chat service Zeta, cutting serving costs by 45%.
Developed Pingpong AI platform, enabling businesses to leverage the companyโ€™s LLM models.
Developed large-scale pseudonymization pipeline for nTBs of service data using Apache Beam, Airflow, and BigQuery.
Optimized infrastructure for AI chatbot โ€˜Luda Leeโ€™ and established a continuous ML learning pipeline.

Gaudio Lab | Software Engineer

2020.12 ~ 2021.07
Developed a real-time, on-device inference engine for audio source separation using TensorFlow Lite and CoreML, implemented in C++.
โ€ข
Reduced real-time inference speed by 25% compared to the previous implementation by leveraging Model Quantization, XNNPack, and NNAPI.
โ€ข
Developed an iOS/Android app to demo C++ On-device Source Separation Engine as a Karaoke Solution.
โ€ข
Deployed and maintained MLOps solutions, such as Kubeflow, MLflow, and DVC, to enhance and support internal research pipelines.

Education

Hanyang Univ. | Department of Computer Science

2019.03 ~

Incheon Science High School

2016.03 ~ 2019.02

Awards & Projects

Final Winner - Prompter Day 2023

OpenAI, SKT, 2023.09.
Built a LLM-based tool for helping people to understand hard-to-read paragraphs on internet.
SKT, ์˜คํ”ˆAI์™€ ํœด๋จธ๋‹ˆํ‹ฐ AI์ธ์žฌ ์œ„ํ•œ โ€˜๊ธ€๋กœ๋ฒŒ AI ํ•ด์ปคํ†คโ€™ ์„ฑ๋ฃŒ... โ€˜์„ ์ฒœ์  ํ”„๋กฌํ”„ํ„ฐ๋“คโ€™ํŒ€ ๋Œ€์ƒ
SKํ…”๋ ˆ์ฝค(๋Œ€ํ‘œ ์œ ์˜์ƒ)์ด ๋ฏธ๊ตญ ์˜คํ”ˆAI(OpenAI)์™€ ํ•จ๊ป˜ ๊ฐœ์ตœํ•œ ๊ธ€๋กœ๋ฒŒ AI ํ•ด์ปคํ†ค โ€˜ํ”„๋กฌํ”„ํ„ฐ ๋ฐ์ด ์„œ์šธ 2023(Prompter Day Seoul 2023)โ€™์„ ์„ฑ๊ณต์ ์œผ๋กœ ๋งˆ๋ฌด๋ฆฌํ–ˆ๋‹ค๊ณ  25์ผ ๋ฐํ˜”๋‹ค.SKT์™€ ์˜คํ”ˆAI๊ฐ€ ๊ณต๋™ ๊ฐœ์ตœํ•œ ์ด๋ฒˆ ํ•ด์ปคํ†ค์€ โ€˜AI๋ฅผ ํ†ตํ•ด ํœด๋จธ๋‹ˆํ‹ฐ(์ธ๋ฅ˜)์— ๋„์›€์ด ๋  ์ˆ˜ ์žˆ๋Š” ์„œ๋น„์Šค ๊ฐœ๋ฐœโ€™์„ ์ฃผ์ œ๋กœ ๊ฑด๊ฐ•ํ•œ AI ์ƒํƒœ๊ณ„๋ฅผ ๋งŒ๋“ค์–ด ๋‚˜๊ฐˆ ๊ธ€๋กœ๋ฒŒ ์ธ์žฌ๋“ค์„ ๋ฐœ๊ตดํ•˜๊ธฐ ์œ„ํ•ด ์ง„ํ–‰๋๋‹ค.โ€˜ํ”„๋กฌํ”„ํ„ฐ ๋ฐ์ด ์„œ์šธ 2023โ€™์€ ์ด 226์—ฌ ํŒ€์ด ์˜คํ”ˆAI์˜ ์ƒ์„ฑํ˜• AI ๋ชจ๋ธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ์ฐฝ์˜์ ์ธ ์•„์ด๋””์–ด๋ฅผ ์ œ์ถœํ•˜๋Š” ๋“ฑ ๋œจ๊ฑฐ์šด ๊ด€์‹ฌ์„

Prize by Minister of Science and ICT (1st Place) - PIDICON 2022

Ministry of Science and ICT, 2022.12.
Personal Information Deidentification Competition
์ธ๊ณต์ง€๋Šฅ ์ด๋ฃจ๋‹ค ํ”Œ๋žซํผ '์Šค์บํ„ฐ๋žฉ', ๊ณผ๊ธฐ์ •ํ†ต๋ถ€ โ€˜2022 ๊ฐ€๋ช…ยท์ต๋ช…์ฒ˜๋ฆฌ ๊ธฐ์ˆ  ๊ฒฝ์ง„๋Œ€ํšŒ' ๋Œ€์ƒ ์ˆ˜์ƒ
AI ์Šคํƒ€ํŠธ์—… ์Šค์บํ„ฐ๋žฉ(๋Œ€ํ‘œ ๊น€์ข…์œค)์€ ๊ณผํ•™๊ธฐ์ˆ ์ •๋ณดํ†ต์‹ ๋ถ€๊ฐ€ ์ฃผ์ตœํ•˜๊ณ  ํ•œ๊ตญ์ธํ„ฐ๋„ท์ง„ํฅ์›(KISA)์ด ์ฃผ๊ด€ํ•œ โ€˜2022 ๊ฐ€๋ช…ยท์ต๋ช…์ฒ˜๋ฆฌ ๊ธฐ์ˆ  ๊ฒฝ์ง„ ๋Œ€ํšŒ'์—์„œ ์ผ๋ฐ˜๋ถ€ ๋Œ€์ƒ์„ ์ˆ˜์ƒํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.์•ˆ์ „ํ•œ ๋ฐ์ดํ„ฐ ํ™œ์šฉ์„ ์œ„ํ•ด ๊ฐ€๋ช…ยท์ต๋ช… ์ฒ˜๋ฆฌ ๊ธฐ์ˆ  ๋ฐœ๊ตด ๋ฐ ์ €๋ณ€ ํ™•๋Œ€๋ฅผ ๋ชฉํ‘œ๋กœ ์ง„ํ–‰๋œ ์ด๋ฒˆ ๋Œ€ํšŒ๋Š” 1์ฐจ ์˜ˆ์„ ์—์„œ ์œ ํ†ต ๋ถ„์•ผ์˜ ์›๋ณธ ์ •๋ณด๋ฅผ ์ต๋ช…์ฒ˜๋ฆฌ ํ•˜๋Š” ๊ธฐ์ˆ ์„ ๊ฒจ๋ค˜๊ณ , 2์ฐจ ๋ณธ์„ ์—์„œ๋Š” ๊ณต๊ณต ํ–‰์ • ๋ถ„์•ผ์˜ ์ •๋ณด๋ฅผ ๊ฐ€๋ช…๊ณผ ์ต๋ช…์ฒ˜๋ฆฌํ•˜๋Š” ๊ฒƒ์€ ๋ฌผ๋ก  ์ ์ •์„ฑ ํ‰๊ฐ€ ๊ธฐ์ค€ ๋ฐ ๋‚ด์šฉ์„ ๋ฐœํ‘œํ•˜๋Š” ๋ฐฉ์‹์œผ๋กœ ์ง„ํ–‰๋˜์—ˆ๋‹ค.์Šค์บํ„ฐ๋žฉ์—์„œ โ€˜์ด๋ฃจ๋‹ค'๋ฅผ ๋งŒ๋“œ๋Š” ๊ฐœ๋ฐœ์ž๋กœ ๊ตฌ์„ฑ๋œ ์ฐธ๊ฐ€ํŒ€(ํŒ€๋ช…

Finalist (Track Winner of Chainapsis) - Junction Asia 2022

SHIFT, 2022.08.
Built open-source community for musical composition above Cosmos SDK and Keplr Wallet from Chainapsis.

3st Place - Azure Virtual Hackathon (DevOps with GitHub Track)

Microsoft, 2022.06
Developed a CI tool for Multi-stage Review Process
castanets
team-castanets

Prize by Minister of Science and ICT - Software Maestro 11th

Minister of Science and ICT, 2021.01.
Built an AI application BANJU, AI assisted chord accompany assistant.