DontApply.Club
Login Register

ML Engineer, Infrastructure

Toss Securities β€” South Korea Β· Posted ~2 weeks ago

Skills

Kubernetes Machine learning infrastructure AI infrastructure GPU cluster operations Infrastructure performance optimization Distributed systems Resource management Performance profiling InfiniBand H100 B300 GPU clusters High-performance storage

πŸ”“ Log in to save this job, tailor your resume & track your apply process β€” 7 days free, no card needed.

Log in to add to target list

Summary ✨ AI‑Generated

An ML Infrastructure Engineer role focused on designing and operating high-performance AI computing environments. Responsibilities include managing Kubernetes-based GPU clusters and high-speed storage, building resource observability and allocation systems, automating infrastructure utilization, and developing profiling environments to identify model performance bottlenecks.

Highlights

Work on high-performance AI infrastructure, large GPU clusters, high-speed storage, automated resource management, observability, and performance optimization at significant scale.

Description

ν•©λ₯˜ν•˜κ²Œ 될 νŒ€μ— λŒ€ν•΄ μ•Œλ €λ“œλ €μš” ν† μŠ€μ¦κΆŒ ML Engineer(Infra)λŠ” Product Division λ‚΄ ML Platform Team에 속해 μžˆμ–΄μš”.ML Platform Team의 λͺ©ν‘œλŠ” ν† μŠ€μ¦κΆŒμ˜ λ‹€μ–‘ν•œ AI/ML μ„œλΉ„μŠ€λ“€μ„ 효율적이고 μ•ˆμ •μ μœΌλ‘œ κ°œλ°œν•˜κ³  μš΄μ˜ν•  수 μžˆλŠ” 졜적의 λ¨Έμ‹ λŸ¬λ‹ ν”Œλž«νΌμ„ λ§Œλ“œλŠ” κ±°μ˜ˆμš”.ML Engineer(Infra)λŠ” νŒ€ λ‚΄μ—μ„œ λŒ€κ·œλͺ¨ AI μΈν”„λΌμ˜ νš¨μœ¨μ„±μ„ κ·ΉλŒ€ν™”ν•˜κ³  λ¦¬μ†ŒμŠ€ μ‚¬μš©μ„ μ •κ΅ν•˜κ²Œ μ œμ–΄ν•˜λ©°, 인프라 μ„±λŠ₯을 μ΅œλŒ€κΉŒμ§€ λŒμ–΄μ˜¬λ¦¬λŠ” 고도화 μž‘μ—…μ— 집쀑할 μ˜ˆμ •μ΄μ—μš”. ν•©λ₯˜ν•˜λ©΄ ν•¨κ»˜ ν•  μ—…λ¬΄μ—μš” μ΄ˆκ³ μ„±λŠ₯ AI μ»΄ν“¨νŒ… ν™˜κ²½μ„ μ•ˆμ •μ μœΌλ‘œ 섀계 및 μš΄μ˜ν•΄μš”.InfiniBand둜 μ—°κ²°λœ H100, B300 μ‹œλ¦¬μ¦ˆ λ“± 졜고 μ‚¬μ–‘μ˜ GPU ν΄λŸ¬μŠ€ν„°μ™€ 400GbpsκΈ‰ κ³ μ„±λŠ₯ μŠ€ν† λ¦¬μ§€λ₯Ό μΏ λ²„λ„€ν‹°μŠ€ ν™˜κ²½μ—μ„œ 섀계 및 μš΄μ˜ν•΄μš”.λ‹¨μˆœνžˆ 인프라λ₯Ό κ΅¬μΆ•ν•˜λŠ” 것을 λ„˜μ–΄, ν•˜λ“œμ›¨μ–΄ μ„±λŠ₯을 λκΉŒμ§€ λŒμ–΄λ‚Ό 수 μžˆλ„λ‘ λ„€νŠΈμ›Œν¬μ™€ μŠ€ν† λ¦¬μ§€λ₯Ό μ΅œμ ν™”ν•΄μš”.AI 인프라 전체λ₯Ό ν•œλˆˆμ— 보고 μ œμ–΄ν•˜λŠ” μ‹œμŠ€ν…œμ„ λ§Œλ“€μ–΄μš”.사내 인프라와 μ™ΈλΆ€ ν΄λΌμš°λ“œμ— λΆ„μ‚°λœ AI μžμ› ν˜„ν™©μ„ ν†΅ν•©ν•΄μ„œ λ³Ό 수 μžˆλŠ” κ΄€μΈ‘ μ‹œμŠ€ν…œμ„ κ΅¬μΆ•ν•΄μš”.νŠΉμ • μ„œλΉ„μŠ€κ°€ μžμ›μ„ λ…μ ν•˜μ§€ μ•Šλ„λ‘ μ œμ–΄ν•˜κ³ , μ€‘μš”λ„μ— 따라 λ¦¬μ†ŒμŠ€λ₯Ό μ •κ΅ν•˜κ²Œ ν• λ‹Ήν•˜λŠ” 관리 κΈ°λŠ₯을 κ°œλ°œν•΄μš”.κ°€μž₯ 효율적인 λ¦¬μ†ŒμŠ€ μ‚¬μš©μ„ μœ„ν•œ μžλ™ν™” 도ꡬλ₯Ό κ°œλ°œν•΄μš”.μ‚¬μš©μžλ“€μ΄ 무심코 λ‚­λΉ„ν•˜λŠ” μžμ›μ΄ 없도둝, μ‹€μ œ μ‚¬μš© νŒ¨ν„΄μ„ 뢄석해 β€˜λ”± λ§žλŠ” λ¦¬μ†ŒμŠ€β€™λ₯Ό μΆ”μ²œν•΄ μ£ΌλŠ” 도ꡬλ₯Ό λ§Œλ“€μ–΄μš”.λͺ¨λΈμ˜ μ‹€μ‹œκ°„ μ„±λŠ₯μ΄λ‚˜ μ—λŸ¬μœ¨μ„ 감지해 μžλ™μœΌλ‘œ 규λͺ¨λ₯Ό λŠ˜λ¦¬κ±°λ‚˜ 쀄이고, ν•„μš”ν•œ 곳에 GPUλ₯Ό μž¬λ°°μΉ˜ν•˜λŠ” κΈ°λŠ₯을 κ΅¬ν˜„ν•΄μš”.λͺ¨λΈ μ„±λŠ₯의 병λͺ©μ„ μ°Ύμ•„ ν•΄κ²°ν•˜λŠ” ν™˜κ²½μ„ μ‘°μ„±ν•΄μš”.λͺ¨λΈ ν•™μŠ΅μ΄λ‚˜ μ„œλΉ™ 쀑에 속도가 λŠλ €μ§€λŠ” ꡬ간이 어디인지 μ •ν™•νžˆ μ°Ύμ•„λ‚Ό 수 μžˆλŠ” ν”„λ‘œνŒŒμΌλ§ ν™˜κ²½μ„ κ΅¬μΆ•ν•΄μš”.ν•˜λ“œμ›¨μ–΄μ™€ μ†Œν”„νŠΈμ›¨μ–΄ μ‚¬μ΄μ—μ„œ λ°œμƒν•˜λŠ” μ„±λŠ₯ μ €ν•˜ 원인을 λΆ„μ„ν•˜κ³  κ°œμ„ ν•  수 μžˆλ„λ‘ μ§€μ›ν•΄μš”. 이런 λΆ„κ³Ό ν•¨κ»˜ν•˜κ³  μ‹Άμ–΄μš” λŒ€κ·œλͺ¨ νŠΈλž˜ν”½μ„ μ²˜λ¦¬ν•˜λŠ” μΏ λ²„λ„€ν‹°μŠ€ 기반 ML 인프라λ₯Ό κ΅¬μΆ•ν•˜κ³  μš΄μ˜ν•΄ λ³Έ κ²½ν—˜μ΄ ν•„μš”ν•΄μš”.λ‹¨μˆœ κ°œλ°œμ„ λ„˜μ–΄, μ‹€μ œ 라이브 μ„œλΉ„μŠ€λ₯Ό μ•ˆμ •μ μœΌλ‘œ μš΄μ˜ν•˜λŠ” 데에 μ±…μž„κ°μ„ λŠλΌλŠ” 뢄이 ν•„μš”ν•΄μš”.μž₯μ• κ°€ λ°œμƒν–ˆμ„ λ•Œ κ·Όλ³Έ 원인을 끈기 있게 λΆ„μ„ν•˜κ³  λ””λ²„κΉ…ν•˜μ—¬ ν•΄κ²°ν•΄ λ³Έ κ²½ν—˜μ΄ ν•„μš”ν•΄μš”.μ‹œμŠ€ν…œ λ¦¬μ†ŒμŠ€(GPU/CPU/Memory/Network/Storage)의 λ™μž‘ 원리λ₯Ό 잘 μ΄ν•΄ν•˜κ³  있고, 이λ₯Ό λͺ¨λ‹ˆν„°λ§ν•  수 μžˆλŠ” μ‹œμŠ€ν…œμ„ ꡬ좕해 λ³Έ κ²½ν—˜μ΄ ν•„μš”ν•΄μš”.μ„œλΉ„μŠ€ 운영 쀑 λ°œμƒν•˜λŠ” λ‹€μ–‘ν•œ λ¬Έμ œλ“€μ„ ν•΄κ²°ν•˜λ©°, μ‹œμŠ€ν…œμ„ 더 κ²¬κ³ ν•˜κ²Œ λ°œμ „μ‹œν‚€λŠ” 과정에 κ°€μΉ˜λ₯Ό λ‘μ‹œλŠ” 뢄이 ν•„μš”ν•΄μš”. 이런 κ²½ν—˜μ΄ μžˆλ‹€λ©΄ 더 μ’‹μ•„μš” λŒ€κ·œλͺ¨ ν΄λŸ¬μŠ€ν„°μ˜ μžμ› μ‚¬μš© ν˜„ν™©μ„ 톡합 κ΄€μΈ‘ν•΄ λ³Έ κ²½ν—˜μ΄ 있으면 μ’‹μ•„μš”.Quota 및 Rate Limit 등을 톡해 λ¦¬μ†ŒμŠ€λ₯Ό μ²΄κ³„μ μœΌλ‘œ μ œμ–΄ν•˜λŠ” μ‹œμŠ€ν…œμ„ ꡬ좕해 λ³Έ κ²½ν—˜μ΄ 있으면 μ’‹μ•„μš”.Kubeflowλ‚˜ Kubernetes λ“± μ˜€ν”ˆ μ†ŒμŠ€ ν”Œλž«νΌμ˜ λ‚΄λΆ€ μ½”λ“œλ₯Ό 깊이 νŒŒμ•…ν•˜κ³ , ν•„μš”μ— 따라 직접 μˆ˜μ •ν•΄μ„œ μ‚¬μš©ν•΄ λ³Έ κ²½ν—˜μ΄ 있으면 μ’‹μ•„μš”.Nsight Systems/Compute, PyTorch Profiler 같은 μ „λ¬Έ 도ꡬλ₯Ό μ‚¬μš©ν•΄ 컀널 λ ˆλ²¨μ—μ„œ 병λͺ©μ„ λΆ„μ„ν•˜κ³  μ΅œμ ν™”ν•΄ λ³Έ κ²½ν—˜μ΄ 있으면 μ’‹μ•„μš”.μ›Œν¬λ‘œλ“œ νŠΉμ„±μ— 맞좰 λΉ„μš©μ„ μ•„λΌκ±°λ‚˜ μ„±λŠ₯을 λ†’μ΄λŠ” 업무(Rightsizing, Cost Optimization)λ₯Ό 직접 섀계해 λ³Έ κ²½ν—˜μ΄ 있으면 μ’‹μ•„μš”.MIG, MPS 같은 GPU 가상화 κΈ°μˆ μ„ λ„μž…ν•΄ μžμ› ν™œμš©λ₯ μ„ κ·Ήν•œμœΌλ‘œ λ†’μ—¬ λ³Έ κ²½ν—˜μ΄ 있으면 μ’‹μ•„μš”. 이λ ₯μ„œλŠ” μ΄λ ‡κ²Œ μž‘μ„±ν•˜μ‹œλŠ” κ±Έ μΆ”μ²œν•΄μš” μž„νŒ©νŠΈ μžˆμ—ˆλ˜ 업무/ν”„λ‘œμ νŠΈμ™€ κ·Έ κ²°κ³Ό(특히 효율 κ°œμ„ , μ„±λŠ₯ μ΅œμ ν™” 수치 λ“±)에 λŒ€ν•΄ ꡬ체적으둜 μ μ–΄μ£Όμ„Έμš”.기술적으둜 μ™ΈλΆ€ κ³΅κ°œκ°€ λ―Όκ°ν•œ 사항일 경우, ν•΄λ‹Ή 뢀뢄은 μ œμ™Έν•΄ μ£Όμ„Έμš”.ν•΄κ²°ν•œ λ¬Έμ œλ“€μ— λŒ€ν•΄ μ–΄λ–€ 방법둠듀을 μ–΄λ–€ 이유둜 μ μš©ν–ˆλŠ”μ§€ μžμ„Ένžˆ μ μ–΄μ£Όμ„Έμš”. ν† μŠ€μ¦κΆŒμ—μ„œ μ‚¬μš©ν•˜λŠ” 기술 Infrastructure: Kubernetes, Kubeflow, Argo CD, HelmCloud & Compute: AWS, Azure, H100/B300 GPU Cluster, Infiniband, NVLink, High-Performance Storage (400Gbps)Serving & Optimization: vLLM, SGLangObservability & Data: Prometheus, Grafana, Elasticsearch, Kafka, DCGM, NsightLanguages: Python ν† μŠ€μ¦κΆŒμœΌλ‘œμ˜ ν•©λ₯˜μ—¬μ • μ„œλ₯˜μ ‘μˆ˜ > 프리 인터뷰 > 직무 인터뷰 > 문화적합성 인터뷰 > 레퍼런슀 체크 > μ²˜μš°ν˜‘μ˜ > μ΅œμ’…ν•©κ²© 및 μž…μ‚¬ κΌ­ 확인해 μ£Όμ„Έμš” 이λ ₯μ„œ 및 제좜 μ„œλ₯˜μ— ν—ˆμœ„ 사싀이 λ°œκ²¬λ˜κ±°λ‚˜ 근무 이λ ₯ 쀑 징계사항이 확인될 경우, μ±„μš©μ΄ μ·¨μ†Œλ  수 μžˆμ–΄μš”.ν† μŠ€μ¦κΆŒ λ‚΄κ·œμ— 따라 μ±„μš© κΈˆμ§€μž λ˜λŠ” κ²°κ²©μ‚¬μœ  ν•΄λ‹ΉμžλŠ” μ±„μš©μ΄ μ·¨μ†Œλ  수 μžˆμ–΄μš”.μž₯애인 및 κ΅­κ°€λ³΄ν›ˆλŒ€μƒμžλŠ” 지원 μ‹œ 관련법에 따라 μš°λŒ€ν•˜κ³  μžˆμ–΄μš”. ν•¨κ»˜ ν•  λ™λ£Œλ₯Ό μœ„ν•œ ν•œλ§ˆλ”” "AI μΈν”„λΌμ˜ 효율과 μ„±λŠ₯을 κ·Ήν•œμœΌλ‘œ λŒμ–΄μ˜¬λ¦΄ λ™λ£Œλ₯Ό μ°Ύκ³  μžˆμ–΄μš”" ν† μŠ€μ¦κΆŒμ€ 이미 B300 λ“± 졜고의 λ¦¬μ†ŒμŠ€λ₯Ό ν™•λ³΄ν•˜κ³ , μˆ˜λ§Žμ€ AI μ„œλΉ„μŠ€λ₯Ό 라이브둜 μš΄μ˜ν•˜κ³  μžˆλŠ” μ‘°μ§μ΄μ—μš”. 이제 μš°λ¦¬λŠ” 'λ¬Έμ œμ—†μ΄ λŒμ•„κ°€λŠ”' μ‹œμŠ€ν…œμ„ λ„˜μ–΄, 'κ°€μž₯ 효율적이고 λΉ λ₯Έ' μ‹œμŠ€ν…œμ„ λ§Œλ“œλŠ” λ‹¨κ³„λ‘œμ˜ 도약을 μ€€λΉ„ν•˜κ³  μžˆμ–΄μš”.ν­λ°œν•˜λŠ” AI μˆ˜μš”λ₯Ό κ°λ‹Ήν•˜λ©΄μ„œλ„ 인프라 νš¨μœ¨μ„ 획기적으둜 κ°œμ„ ν•˜κ³ , 보이지 μ•ŠλŠ” 병λͺ©κΉŒμ§€ μ§‘μš”ν•˜κ²Œ μ°Ύμ•„λ‚΄ μ΅œμ ν™”ν•˜λŠ” 도전적인 κ³Όμ œλ“€μ΄ μ—¬λŸ¬λΆ„μ„ 기닀리고 μžˆμ–΄μš”. λŒ€κ·œλͺ¨ ν΄λŸ¬μŠ€ν„°μ˜ ν˜„ν™©μ„ ν•œλˆˆμ— νŒŒμ•…ν•˜κ³ , 기술적 ν•œκ³„λ₯Ό λ„˜μ–΄ 졜고의 μ„±λŠ₯을 λ§Œλ“€μ–΄λ‚΄λŠ” μ§œλ¦Ών•œ κ²½ν—˜μ„ μ›ν•˜μ‹ λ‹€λ©΄ μ§€κΈˆ λ°”λ‘œ ν† μŠ€μ¦κΆŒμ— ν•©λ₯˜ν•˜μ„Έμš”!