典型案例

Seongeun So

Key takeaway: 本文为资深AI技术专家Seongeun So的LinkedIn职业档案全览。Seongeun So拥有超过13年行业经验,当前担任OpenAI的AI Deployment Engineer。此前,他在Google担任AI Forward Deployed Engineer (FDE),专注于日本及亚太地区(JAPAC)的AI模型实战部署与技术落地。其职业生涯横跨全球顶尖科技巨头与本土初创公司:他曾任Amazon Web Services (AWS)解决方案架构师,为全球公共部门客户提供AI/ML上云咨询;创立的AdTech公司Remake Digital获得Kakao ventures等投资,并担任CTO开发了自动化广告设计引擎Crolo和Databrush;在NAVER Corp作为研究工程师期间,他主导了CLOVA AI音箱的对话系统研发,并开发了使用视频目标检测追踪技术的KBO本垒打识别系统;早期还在SELVAS AI Inc.从事手写字符识别(OCR)、LSTM及CNN等经典机器学习算法的研究。Seongeun So毕业于光州科学技术院(计算机科学硕士)及首尔科学技术大学(电气工程学士)。其近期在LinkedIn发布的动态显示,他对RAG系统的幻觉缺陷、开源代码债务、LLM安全护栏及AI代理的‘自主研究’(autoresearch)等前沿话题有深入洞察,展现了从底层算法研发到顶级商业落地的完整FDE能力模型。

Key Takeaways
  1. Seongeun So于2026年7月从Google AI FDE岗位离职,加入OpenAI担任AI Deployment Engineer。
  2. 其在Google任职1年9个月,专注于JAPAC大区的AI模型前线部署工程,是典型的FDE实践者。
  3. 曾作为首位工程师加入Remake Digital并担任CTO,主导开发Crolo和Databrush引擎,公司最终被成功收购。
  4. 在NAVER Corp参与CLOVA AI音箱的对话管理及语言生成模块研发,并开发了视频分析系统(如KBO本垒打检测)并成功上线。
  5. Seongeun So近期频繁分享关于RAG系统幻觉、开源Fork策略带来的技术债务及AI代理自主研究范式的专业博客。
  6. 其职业路径完整覆盖了机器学习、NLP、计算机视觉以及大规模AI系统部署,体现了从学术研究到产业落地的闭环。
在AI行业,从‘研究’到‘落地’之间一直存在一条巨大的鸿沟。Seongeun So的这份职业档案,恰好为‘AI部署工程师’以及近年大火的‘前线部署工程师’这两个神秘职位,提供了一份教科书式的履历画像。他并非单纯的后端运维,而是贯穿了NLP、计算机视觉、对话系统的全栈研发,既能在NAVER打造日活百万的AI音箱系统,也能在AWS架构云端方案,甚至亲自下场当CTO从零搭建技术团队。他从Google FDE到OpenAI AI Deployment Engineer的职业跃迁,清晰揭示了顶尖AI实验室对于‘能将模型转化为商业价值’的复合型人才的极度渴求。对于所有好奇‘AI落地最后一公里’究竟需要何种能力的从业者来说,这绝对是一份值得逐行细读的参考范本。

Seongeun So于2026年7月从Google AI FDE岗位离职,加入OpenAI担任AI Deployment Engineer。

—— 络石智能编辑部 · Editor's Pick

Seongeun So

AI Deployment Engineer @ OpenAI

Seoul, South Korea (KR)

500 connections • 3,313 followers

About

Currently, I am working at OpenAI as an AI Deployment Engineer, where I focus on helping…

Total Experience: 13 years and 10 months

Experience

AI Deployment Engineer - OpenAI (Current)

Jul 2026 - Present (1 month) in Seoul, South Korea

Department: Engineering and Technical

OpenAI is a Research Services company. OpenAI is a company that believes in building safe and beneficial AGI (Artificial General Intelligence) systems to solve human-level problems. OpenAI has 6,000-7,000 employees (+60% YoY) and has annual revenue in the $1B-$2B range, founded in 2015. Headquartered in San Francisco, California, United States. Its workforce is distributed across 88 countries (including United States, United Kingdom, India, Ireland, and Japan). Has $196B in total funding, with 19 prior funding rounds. Research output of 1,000-2,000 works, primarily in Physics and Astronomy, Computer Science, and Social Sciences. Funds research behind 1-10 works.

AI Forward Deployed Engineer - Google

Oct 2024 - Jul 2026 (1 year and 9 months) in Seoul, South Korea

▪ AI FDE @ JAPAC Forward Deployed Engineering team

Department: Engineering and Technical • Level: Specialist

Google (GOOGL) is a Software Development company. Google is a technology company that offers a wide range of services and products to help businesses grow. They provide engineering jobs, artificial intelligence and machine learning, consumer hardware design and build systems, platforms and ecosystems, site reliability engineer, teams, engineering and tech, sales, service and support, and people. Google has 100,000-200,000 employees (1-2% YoY decline) and has annual revenue in the $300B-$400B range, founded in 1998. Headquartered in Mountain View, California, United States, operates in 17 countries (including Australia, Spain, France, Italy, and Switzerland). Its workforce is distributed across 201 countries (including United States, India, United Kingdom, Ireland, and Canada). Has a market cap in the $500B-$600B range. IPO 2004.

Solutions Architect - Amazon Web Services (AWS)

Feb 2022 - Oct 2024 (2 years and 8 months) in Seoul, South Korea

▪ Solutions Architect at World Wide Public Sector ▪ Helps customers use AI/ML technology on AWS

Department: Engineering and Technical • Level: Specialist

Amazon Web Services (AWS) is a IT Services and IT Consulting company. Amazon Web Services (AWS) has 100,000-200,000 employees (+2-3% YoY) and has annual revenue in the $100B-$200B range, founded in 2006. Headquartered in Seattle, Washington, United States, operates in 8 countries (including Japan, Luxembourg, Brazil, United Kingdom, and Italy). Its workforce is distributed across 159 countries (including United States, India, United Kingdom, Canada, and Ireland).

CTO - Remake Digital

Nov 2019 - Feb 2022 (2 years and 3 months) in Seoul, Korea

▪ Digital marketing and AdTech startup funded by Kakao ventures, Fast Investment and SNOW Corp. ▪ As a CTO and the first engineer of the company, I created service named Crolo (crolo.io) and Databrush (databrush.io), which designs creatives for advertisements automatically. Focused on Engine part, which generates image banner based on lots of algorithms like Engineering, Computer Vision, Natural Language Processing, and so on. ▪ I do research and develop, as well as hiring and leading R&D team. Also created development culture and entire dev process. ▪ Left the company after acquisition.

Department: Engineering and Technical • Level: C-Level

Remake is a Software Development company. Remake has 1-10 employees, founded in 2018. Headquartered in Seoul, Seoul, South Korea. Has $431.5K in total funding.

Research Engineer - NAVER Corp

Feb 2016 - Nov 2019 (3 years and 9 months) in Seongnam, Gyeonggi-do, Korea

▪ As a research engineer, I researched about Natural Language Processing, especially Dialogue System which used for CLOVA AI speaker. Focused on Dialogue Management and Language Generation module. Widely used in Korea and Japan. ▪ Also I developed a Video Analysis system using various algorithms like video object detection with tracking, video action recognition, video classification, captioning, STT and lots of mixed NLP and CV techniques. One of system (Homerun detection for KBO) is released to real service and was very successful.

Department: Research • Level: Specialist

NAVER Corp (035420.KS) is a Technology, Information and Internet company. NAVER Corp. is a company that connects the world's all possibilities through technology and services. NAVER Corp has 3,000-4,000 employees (+7-8% YoY) and has annual revenue in the KRW12000B-KRW13000B range, founded in 1999. Headquartered in South Korea, with an office in Seongnam. Its workforce is distributed across 21 countries (including South Korea, China, United States, Japan, and Vietnam). Has a market cap in the KRW30000B-KRW31000B range. IPO 2008.

Research Engineer - SELVAS AI Inc.

Jan 2014 - Feb 2016 (2 years and 1 month) in Seoul, Korea

▪ As a research engineer, researched about Handwritten Character Recognition, Handwritten Math Equation Recognition and OCR. ▪ Most of the system based on LSTM and CNN, also used lots of classical machine learning algorithms like Random Forest and SVM. I improved the performance of engine a lot, and system is widely used by many companies.

Department: Research • Level: Specialist

Software Engineer - NextVerb, Inc

Nov 2011 - Feb 2013 (1 year and 3 months) in Mountain View, CA, USA

▪ Social-based place recommendation service, angel funded startup. ▪ Joined as a co-founder. Experienced about various tasks for Startup like funding, investment, conference presentation / pitching, organizing team and so on. ▪ I created Android application called allTomto. It recommends places based on Facebook data and Google maps location information. We used many social features, collaborative filtering algorithm and clustering POIs to get a good recommendation performance.

Department: Engineering and Technical

Education

Master's Degree Computer Science at Gwangju Institute of Science and Technology

2011 - 2014 (3 years)

BS Electrical at Seoul National University of Science and Technology

2007 - 2011 (4 years)

Skills

adtech • ai (artificial intelligence) • ai deployment • ai prototyping • ai/ml • algorithms • analysis • android • android application development • android development • apac • architect • arm • artificial intelligence • aws • banner • big data • c • c++ • classical machine learning • clustering • collaboration • collaborative filtering • communication • communication framework • computer vision • conference presentation • creatives • cto • data • data mining • deep learning • deployment • deployment engineering • designs • detection • dev • development • dialogue management • dialogue system • digital marketing • education • education research • embedded systems • emulator • engineer • engineering • facebook marketing • fast • fde • forward deployed • forward deployed engineer • forward deployed engineering • framework • fsa • generative ai • genetics • get • google maps • hadoop • hadoop framework • hiring • investment management • java • kakao • language processing • leading • learning algorithms • lstm • machine learning • machine learning algorithms • machine learning techniques • management • maps • marketing • math • mining • mobile applications • mobile development • mobile devices • modeling • mongodb • natural • natural language • natural language generation • natural language processing • nlp • node.js • object detection • optical character recognition (ocr) • organizing • perl • presentation • processing • programming • prototyping • python • r&d • random forest • research • rnn • robust communication • service • servicenow • software development • software engineering • solutions architect • startup • stt • subversion • system engineer • technology • video analysis • vulnerability • vulnerability analysis • well • windows • windows mobile

Languages

Korean - Native or bilingual proficiency

English - Professional working proficiency

Publications

Total of 1 work and 140 citations, including 1 work in 2013, with an h-index of 1.

  • Jeong-Kyun Kim, Seongeun So, Heejin Lee, et al. DigSee: disease gene search engine with evidence sentences (version cancer). Nucleic Acids Research. 2013. doi:10.1093/nar/gkt531. — 140 citations

Research areas

Research areas include Bioinformatics and Genomic Networks, Biomedical Text Mining and Ontologies, and Machine Learning in Bioinformatics.

Frequent collaborators

  • Jung‐Jae Kim — Agency for Science, Technology and Research
  • Jeong-Kyun Kim — Agency for Science, Technology and Research
  • Hyunju Lee — Chonnam National University
  • Heejin Lee — Kookmin University
  • Jong C. Park — Korea Advanced Institute of Science and Technology
  • Hee-Jin Lee — Korea Advanced Institute of Science and Technology

Publication profile

  • Most cited recent work: DigSee: disease gene search engine with evidence sentences (version cancer) (56 recent citations)
  • Top venues: Nucleic Acids Research
  • Languages: en (1)

Tags

Related Topics

Expert Comment

This article is curated by the editorial team from public sources for reference only.

FAQ

Seongeun So为何从Google跳槽到OpenAI,他的职业发展路径是怎样的?
Seongeun So于2026年7月从Google的AI Forward Deployed Engineer岗位离职,加入了OpenAI担任AI Deployment Engineer。他的职业路径具有典型的‘全栈落地’特征:早期在NAVER、SELVAS积累了深厚的NLP和计算机视觉研发经验,后在AWS从事解决方案架构师拓展了行业视野,并有过CTO的创业经历,最终进入Google和OpenAI等顶级AI巨头从事前线模型部署。
什么是AI Forward Deployed Engineer (FDE),其主要职责是什么?
AI Forward Deployed Engineer (FDE) 是连接尖端AI研发与客户实际需求的桥梁。以Seongeun So在Google的工作为例,其核心职责不是坐在办公室写代码,而是深入JAPAC(日本及亚太)等区域,直接面对客户,将谷歌的AI模型和工具集成到客户的实际业务流程中,解决真实世界的工程难题。这要求FDE同时具备深厚的算法背景、强大的系统架构能力以及优秀的客户沟通与项目交付能力。
Seongeun So在AI技术领域有哪些具体的实践经验和代表项目?
Seongeun So在AI领域拥有多项具体的实践成果:在NAVER期间,他参与了CLOVA AI智能音箱的对话管理和语言生成模块开发,并研发了应用于KBO(韩国职业棒球联赛)的本垒打视频分析识别系统;创立Remake Digital期间,他主导开发了自动化广告创意设计引擎Crolo和Databrush;此外,他还关注LLM应用中的具体工程问题,例如RAG系统在缺少防护时的幻觉漏洞,以及开源项目Fork策略带来的长期技术债务等。

Related Articles

FDE前沿部署工程师实战指南:从模型部署到AI Agent系统构建

本文是一份面向前沿部署工程师(FDE)的完整实战指南,系统拆解了 FDE 的核心内涵、技能树、项目实战和学习路径。文章指出 FDE 不是简单的“模型部署”,而是确保复杂 AI 能力(大语言模型、多模态模型、AI Agent、RAG)能够在真实生产系统中稳定、高效、可扩展且低成本地集成的全栈角色,其 40k 以上月薪是对“AI 应用最后一公里”复杂性的定价。核心技能体系覆盖 AI 基础(Transformer、Prompt 工程、RAG、Agent)、工程开发(Python、FastAPI 异步编程)、云原生(Docker、Kubernetes、Istio、Prometheus/Grafana)和系统设计。实战部分通过使用 vLLM 部署 Qwen2.5-7B-Instruct-AWQ 量化模型、构建 FastAPI 代理网关实现路由与监控、编写 Dockerfile 和 Kubernetes Deployment 文件完成容器化与编排,以及部署多智能体项目 My AI Town 来串联所有技能。文章还提供了常见故障排查清单(超时、OOM、推理慢、Agent 循环、Pod 重启)、最佳实践(IaC、配置分离、可观测性、成本优化、降级策略)及三阶段学习路径(基础巩固、核心技能、项目实战),并解析了面试考察的系统设计、故障排查和工程协作等方向,为求职者提供了从零到就业的路线图。

Read More

FDE 职业路线图:如何成为一名前沿部署工程师

本文是一份指导如何成为一线部署工程师的职业路线图,发表于2026年8月15日。文章明确定义Forward Deployed Engineer是嵌入客户现场的软件工程师,负责将公司的产品适配到该客户的特定环境中,编写生产级代码,并直接与用户沟通,融合了软件工程、模糊问题解决和客户沟通三大能力。路线图分为四个阶段:0-6个月建立坚实的工程基础,主攻Python、辅以Java/JavaScript/Go,掌握REST API、Git和生产级测试;6-18个月增加数据集成与AI部署技能,包括SQL、系统集成、检索管线、嵌入和云基础;18-36个月培养客户判断力,能将模糊需求转化为可交付系统,并对客户结果负责;三年以上进入高级FDE、技术解决方案架构师或VP方向,可选择深耕医疗、金融、制造等行业垂直领域或管理FDE团队。文章指出该角色最初由Palantir普及,现OpenAI、Anthropic、Google、Scale AI等头部AI及基础设施公司均组建FDE团队,市场需求快速扩大。薪酬普遍高于同级软件工程师,但受公司、级别和地区影响较大。转行背景涵盖软件工程师、解决方案顾问、数据分析师和行业专家等,但都需要最终具备相同的技能组合。文章最后提供了FDE就绪检视清单,并对比了FDE与传统解决方案工程师、常规软件工程师的区别。

Read More

什么是前部署工程师?

前线部署工程师(FDE)是一种嵌入客户或操作环境的软件工程师,负责从工作流范围确定、代码构建、系统集成、部署到稳定移交的全过程,并对生产环境结果拥有直接所有权,而非仅提供咨询。文章引用 Alvarez & Marsal 2026 年报告,指出每月 FDE 职位发布量从 2020 年 9 月的不足 10 个增至 2025 年 9 月的近 50 个,反映了企业在 AI 采纳中从模型访问转向实际部署的需求。文章详细区分了 FDE 与软件工程师、解决方案架构师的差异:FDE 在客户环境中构建并维护系统,成功标准是客户工作流改善和可移交性,而非仅满足内部规格。2026 年 FDE 基本工资中位数约 18 万美元(范围 13.5 万至 21.5 万美元),85% 的职位包含股权。核心技能涵盖工程基础、客户现场执行、集成部署判断以及 AI 时代的安全与治理纪律。文章提供了完整的招聘方法,包括定义部署环境、筛选交付证明、实践任务设计以及 19 个面试问题,并介绍了 GoGloby 如何通过嵌入资深 AI 解决方案架构师、安装 Agentic SDLC 和 AI 开发智能层,帮助成熟软件公司安全采用 AI 并完成生产级部署。FDE 角色源于 Palantir,现已被 OpenAI、Google、AWS 等主流企业采纳,其增长的根源在于弥合生成式 AI 巨额投资与 95% 组织未获回报之间的执行差距。

Read More

OpenAI放大招:企业级AI平台Frontier上线,专治各种智能体

OpenAI于2026年2月5日正式推出企业级AI平台Frontier,定位为帮助企业统一构建、部署和管理AI智能体的系统,旨在解决企业AI部署中的“碎片化”和“AI孤岛”问题。Frontier通过连接企业数据仓库、CRM、工单工具等孤立系统,为AI智能体提供共享业务背景、执行环境、记忆与质量优化机制,并内置严格的身份、权限与边界管控,可适用于金融和医疗等受监管领域。平台具有高度开放性,不仅支持OpenAI自建智能体,还兼容来自Google、Microsoft及直接竞争对手Anthropic的第三方智能体,采用开放标准避免供应商锁定。OpenAI应用业务CEO Fidji Simo将智能体视为“AI同事”,预测到2026年底领先企业的大部分数字化工作将由人类指挥的智能体舰队执行。企业业务总经理Barret Zoph指出Frontier创建了“语义层”以打通工作流程。商业层面,OpenAI企业客户约占业务40%,CFO Sarah Friar预计年底接近50%,全球有超过100万家商业客户。平台已面向惠普、Intuit、甲骨文、State Farm、Thermo Fisher和优步等有限客户开放,BBVA、思科和T-Mobile等也在试点。发布前后PayPal、Expedia等软件股大跌,市值蒸发超3000亿美元。OpenAI通过前向部署工程师与客户并肩工作,将实施经验产品化。Frontier的推出标志着AI企业应用从单点工具迈向系统化数字员工管理的新阶段。

Read More

100亿美元的前端部署工程热潮 | Tomasz Tunguz

Tomasz Tunguz 分析了 AI 行业在“前沿部署工程师(FDE)”领域的巨额资本投入。文章指出,在短短 12 个月内,OpenAI、Microsoft、Anthropic、Amazon 和 Google Cloud 等顶级 AI 公司已承诺投入 97.5 亿美元,将软件工程师派驻客户现场以推动 AI 落地。这笔资金的规模相当于埃森哲(Accenture)年劳动力成本的四分之一,标志着 FDE 模式正从 Palantir 的标志性做法转变为全行业标准。 当前市场存在三种主要部署模式:一是微软(25亿美元)和亚马逊(10亿美元)采用的“资产负债表”模式,利用现有团队和预算;二是 OpenAI(40亿美元)与 Anthropic(15亿美元,投资方有 Blackstone、Hellman & Friedman、高盛)借助私募股权组建的“独立法人”模式,OpenAI 为此收购了拥有 150 人的咨询公司 Tomoro;三是 Google Cloud 推出的 7.5 亿美元“合作伙伴生态”模式,资助系统集成商进行部署。 这一浪潮背后的核心驱动力是部署瓶颈。MIT 的“GenAI Divide”报告显示,尽管企业在 2025 年投入了 6840 亿美元,但 95% 的企业 AI 试点项目未能产生可衡量的利润影响,表明模型能力(如 GPT-4、Claude)已足够,但多数企业缺乏落地操作能力。FDE 因此成为强大的商业护城河,它们通过培训锁定客户、挖掘私有的工作流与数据模式来优化模型,并切实提高客户的替换成本,从而推动 AI 厂商实现跨组织的深度防御。

Read More