4 days ago
Seoul, Korea, SouthStaff+
Responsibilities
- Own the technical direction, architecture, roadmap, and operating standards for the inference fleet and control plane.
- Design and operate model serving, orchestration, request routing, capacity and model placement, caching, autoscaling, and multi-tenant policies.
- Build and lead the team, including hiring and development, and coordinate AI Research, IT, and business squads.
- Establish model onboarding, regression testing, release automation, canary deployment, rollback, API gateway authentication, quotas, rate limiting, and token-level observability.
- Define workload-specific SLOs and error budgets covering latency, availability, error rate, and throughput.
- Create operating practices for on-call response, incident management, postmortems, capacity events, and deployment safety.
- Design GPU scheduling, allocation, tenant isolation, quotas, priorities, capacity planning, and GPU cost visibility.
- Apply and tune continuous batching, prefix caching, KV cache offloading, prefill-decode separation, request scheduling, and model routing.
- Improve routing, scheduling, cache, and model integration code for production workloads.
- Optimize agentic and real-time voice workloads, including end-to-end latency and STT–LLM–TTS serving performance.
Requirements
- 10+ years of experience in infrastructure, SRE, or platform engineering for large-scale traffic services, with leadership of teams responsible for critical-path production systems.
- Experience evaluating and selecting serving engines, orchestration systems, and storage layers while explaining performance, cost, and operational trade-offs.
- Experience leading direct architecture and code reviews, defining latency, availability, throughput, and cost metrics, and delivering measurable improvements.
- Experience negotiating SLOs and costs with business organizations and executives and documenting organizational and technical decisions.
- Experience designing and operating deployment, load balancing, scheduling, autoscaling, observability, on-call, and incident response for large-scale distributed systems in Kubernetes or an equivalent orchestration environment.
- Experience deploying and operating LLM or ML model serving, or comparable high-compute, low-latency real-time systems, in GPU production environments.
- Experience operating capacity planning, quotas, priorities, and resource isolation under rapidly changing demand and constrained resources.
- Production programming experience in Python or Go and experience reading and modifying large open-source systems.
- Preferred experience operating vLLM, SGLang, TensorRT-LLM, Triton Inference Server, NVIDIA Dynamo, llm-d, KServe, Ray Serve, or similar inference systems.
- Preferred experience with multi-node GPU environments, NVLink, InfiniBand/RoCE, GPU Operator, DCGM, and MIG.
- Preferred experience serving real-time STT–LLM–TTS, voice, video, or multimodal systems and optimizing WebSocket or SSE pipelines.
- Preferred experience operating agent workloads, LLM gateways such as LiteLLM, model routing layers, FinOps or GPU chargeback/showback, external model APIs, or non-NVIDIA accelerators.
Benefits
- Professional contract employment with a 5-month probationary period.
- Work location: Yeoksam Centerfield.
- The hiring process may include document screening, phone interview, assignment, personality assessment, technical interview, and culture-fit interviews; steps may vary by position and candidate.
- Employment-protected applicants, including individuals with disabilities and national merit beneficiaries, receive preference according to applicable law.
Tech Stack
Categories
About KRAFTON
KRAFTON is a Seoul-based, KRX-listed video game company that develops and publishes PC, console, and mobile titles. Its studios include PUBG Studios, Bluehole Studio, Striking Distance Studios, and Unknown Worlds, best known for the PUBG: Battlegrounds franchise (and PUBG Mobile with Tencent) and games like Subnautica and TERA. The company monetizes through premium sales and live-service content across global markets.