AI Frontier Division - AI Inference Platform & Operations Lead (10+ Years)

 

Introducing our team (project). 

• The AI Frontier division is an organisation that connects KRAFTON's AI technology to real products and services, and validates and scales new business opportunities. Several Squads pursue business for different customers and markets, and each Squad is responsible for everything from product development to service operations and business results. Based on customer needs and actual usage results, we improve products and scale services whose business viability has been confirmed.
• The AI Inference Platform & Operations team builds the inference foundation that these products commonly use, and on that basis builds and operates the inference services provided to external customers. The team's mission is to connect proprietary models and open-weight models to services quickly and reliably, maintain consistent response speed and availability as traffic grows, and operate GPU resources and inference costs efficiently.
• Members of the AI Frontier division share their opinions regardless of rank or years of service, and once a decision is made, they execute quickly. When failures occur, we do not hide them; we use postmortems to make the system more robust, and we document architecture, experiment results and decision-making processes. We use AI tools as everyday development and operations tools, and we aspire to a culture where, when new models and serving technologies emerge, we try them right away and look for ways to apply them in production.

 

Introducing the missions you will take on with our team.

• Own the inference fleet and control plane
  • As the Lead responsible for the technical direction and key technical decisions of the inference platform, you will design the serving platform for proprietary models and open-weight models (including speech and multimodal), establish the build and operation framework, and lead the team's execution.
  • You will select the serving engine and orchestration layer, and establish the technical roadmap and operating standards, including request routing, capacity, model and cache placement, autoscaling, and multi-model and multi-tenant operating policies.
    
• Build the organisation and define the collaboration structure
  • You will directly recruit and develop the team, and serve as the single point of contact between the AI Research compute optimisation R&D organisation, the internal IT organisation that operates the inference services, and each business Squad.
  • You will turn production workload profiles and bottlenecks into R&D requirements, and create a loop for validating, accepting and deploying research results.
•  Safe model onboarding and platform experience
  • You will build performance and quality regression testing for new models, release automation, canary and rollback, authentication, quotas and rate limiting in the API gateway, and token-level observability.
  • You will provide self-service tools and guides so that business Squads and model developers can onboard models quickly and consistently.
• Establish SLOs and operational reliability
  • Together with each Squad, you will set workload-specific SLOs and error budgets, including TTFT, TPOT, p95 and p99 end-to-end latency, availability, error rate and throughput.
  • You will build an operating framework that includes on-call, incident response, postmortems, capacity event response and deploy safety.
  • You will lead postmortem-based architecture improvements.
• Manage GPU resources and unit costs
  • You will design and establish GPU scheduling and allocation, tenant isolation, quota and priority policies, and capacity planning models.
  • You will measure GPU utilisation, throughput, and cost per request, token and job by model and workload, and provide metrics so that each Squad can plan its demand and costs.
• Apply and tune serving techniques and improve the serving system
  • You will establish application strategies for continuous batching, prefix caching, KV cache offloading and tiering, Prefill-Decode separated topologies, request scheduling and model routing that fit real production traffic, and lead the team's execution.
  • The team will directly improve the routing, scheduling, cache logic and model integration code that the service requires.
  • For features provided by the engine or the R&D organisation, such as quantised checkpoints and speculative decoding, you will establish validation criteria for quality regression, performance effects and failure scope, and lead phased deployment.
• Streamline agentic and real-time speech workloads
  • For agents with long contexts and many repeated tool calls, you will provide session-based routing, cache reuse, structured output and cascading between small and large models, and apply them together with the Squads.
  • For speech services, based on the latency and quality targets agreed with the Squads, you will be responsible for the performance and stability of the STT-LLM-TTS inference segments and common serving functions, establish observation criteria for end-to-end latency including turn-taking, and lead improvements.
    
    

We want to grow with someone who has these experiences! (Required)

• Lead role
  • More than 10 years of experience in infrastructure, SRE or platform engineering for large-scale traffic services (internet services, games, platforms, etc.), and experience leading a team responsible for critical-path production systems
  • Someone who can directly compare and select core technologies such as serving engines, orchestration and storage layers, and explain trade-offs between performance, cost and operational complexity with supporting evidence
  • Someone who directly designs and reviews critical technical bottlenecks, leads the team's execution, defines latency, availability, throughput and cost metrics, makes decisions between conflicting goals, and has produced quantitative improvements
  • Someone who can agree on SLOs and costs in numbers with business organisations and executives, and document inter-organisational roles and technical decisions in clear documents
• Common technical base
  • Someone who has designed and operated deployment, load balancing, scheduling, autoscaling, observability, and on-call and incident response frameworks in large-scale distributed systems on Kubernetes or an equivalent orchestration environment
  • Someone who has deployed and operated LLM or ML model serving, or equivalent high-compute, low-latency real-time systems such as real-time 3D games, in production GPU environments
  • Someone who has actually operated capacity planning, quota and priority, and resource isolation policies in environments where sudden demand changes coexist with limited resources
  • Someone who has written production service code in Python or Go, and has read large open-source system code and modified the necessary parts
    
• Eligible for overseas business travel

 

If you have these experiences, you are the person we are looking for! (Preferred)

• Inference platform and GPU operations
  • Experience operating LLM serving engines such as vLLM, SGLang, TensorRT-LLM and Triton Inference Server in production
  • Experience operating distributed inference orchestration or inference control planes such as NVIDIA Dynamo, llm-d, KServe and Ray Serve
  • Experience introducing serving features such as continuous batching, Prefill-Decode separation, prefix caching, and KV cache offloading and tiering into real traffic, and measuring their effects
  • Experience applying optimisation features such as quantised checkpoints and speculative decoding to production together with accuracy regression testing, canary and rollback
  • Experience operating multi-node GPU environments (NVLink, InfiniBand/RoCE) and GPU resource management tools such as GPU Operator, DCGM and MIG
• Real-time speech and agents
  • Experience with speech conversation systems combining STT-LLM-TTS and a Turn Detector, or with real-time streaming serving of speech, video or multimodal models
  • Experience optimising low-latency WebSocket and SSE-based pipelines, including end-to-end TTFT, turn-taking and backpressure
  • Experience serving agent workloads with many tool-calling loops, such as coding agents, or operating an LLM gateway (such as LiteLLM) and a model routing layer
• Organisation, platform and cost
  • Experience building and recruiting and developing a new team or platform organisation from scratch
  • Experience integrating models and algorithms created by researchers into production, leading interface agreement, validation loops, regression testing and deployment
  • Experience providing models to external customers in API form and operating SLAs, quotas and billing
  • Experience creating GPU cost visibility frameworks such as FinOps and chargeback/showback
  • Experience with forking and upstream synchronisation of large open-source systems, or operating release and branch strategies in B2B multi-tenant environments
  • Experience operating serving on non-NVIDIA accelerators such as NPUs

 

---

To take on KRAFTON's challenges, the following selection process is required.

• Document screening > Preliminary interview (Phone Interview) > *Assignment > Personality Assessment > Technical Fit Interview > Comprehensive interviews (2-3 rounds) (Culture Fit Interview) > Offer and joining
• *This process may vary depending on each position and candidate.
• This is an open recruitment and may close early once the position is filled.
• Interviews and job tests may be added if necessary.
• Details of the selection schedule and the pass or fail results will be notified individually by email or phone to the address or number you registered in your application.

 

Please check the required documents!

• Application form (free format), cover letter, career description, portfolio (if required)
• For new graduates, please focus on the cover letter; for experienced candidates, please focus on the career description.
• When attaching a portfolio, please check the guidance below. 
  

 

Work location

• Yeoksam Centerfield

 

Employment type 

• Professional contract