Algorithm - AI System Researcher

FuriosaAI
Seoul
Workplace: OnsiteFull timeFunction: Research & Scientific (R&D)Education: mastersSkills: ["Independent problem structuring","Hypothesis-driven research","Data-driven analysis","Technical communication","Presentation"]

Research and validate next-generation NPU serving approaches, from analytical modeling and roofline/cost analysis to simulator work and real device verification. Explore design-space tradeoffs for cluster-level serving, heterogeneous computing, scheduling/batching, and KV cache or disaggregation strategies. Turn promising ideas into POC/prototypes and partner with the SW organization to help productize results that guide decisions across performance and TCO.

Loading

Loading job details...

Preparing the role view and application actions.

FursaFursa
FuriosaAI
FuriosaAI
1 week ago

Algorithm - AI System Researcher

✓ Verified Job

Canonical indexed version, validated from employer's careers page.

Source: Company careers pageValidated by: Fursa AI
Last checked: 12 minutes agoStatus: Live
Reposted: similar role first listed 4 weeks ago

Job Summary

Research and validate next-generation NPU serving approaches, from analytical modeling and roofline/cost analysis to simulator work and real device verification. Explore design-space tradeoffs for cluster-level serving, heterogeneous computing, scheduling/batching, and KV cache or disaggregation strategies. Turn promising ideas into POC/prototypes and partner with the SW organization to help productize results that guide decisions across performance and TCO.
Location: Seoul
Workplace: Onsite
Employment Type: Full time
Job Function: Research & Scientific (R&D)

Key Responsibilities

  • •Identify major technical challenges for next-gen NPU serving (e.g., cluster-level serving and heterogeneous computing systems) and define them as research tasks.
  • •Explore the design space of serving architectures/algorithms using analytical modeling, roofline analysis, cost models, and an in-house simulator to build quantitative evidence for key decisions.
  • •Design serving algorithms such as scheduling, batching, KV cache management, and disaggregation; validate via modeling/experiments and NPU-based implementations.
  • •Develop validated ideas into POC/prototypes and collaborate with the SW organization to complete productization.
  • •Research across the full cycle from hypothesis setup through real-device validation, balancing analytical modeling and direct implementation as needed.

Key Requirements

  • •MSc (or in-progress) in AI/ML, Computer Architecture, or a related field, or equivalent experience.
  • •Understand the operating principles and performance characteristics of LLM inference (attention, KV cache, prefill/decode, batching, multi-chip parallelism, etc.).
  • •Use Python and/or PyTorch to perform performance modeling, simulation, and data-driven analysis.
  • •Independently structure ambiguous problems and lead hypothesis–validation research cycles.
  • •Communicate analysis results and technical decisions clearly via documents and presentations.
Experience:AI/MLLLM inferenceDistributed systemsCluster-level servingGPU/NPU optimization
Education:Master's
Skills:Independent problem structuringHypothesis-driven researchData-driven analysisTechnical communicationPresentation
Languages:English
Tech Stack:PythonPyTorchVLLMSGLangTensorRT-LLM

Company Brief

FuriosaAI
Designs and develops data‑center AI inference accelerators (RNGD) and a full stack hardware‑software platform to deliver energy‑efficient, high‑performance AI compute for enterprise and cloud customers.
Industry: Hardware Devices
Company Size: Medium (51 to 250 employees)
Growth: Scaleup
Valuation: USD 500M to 1B
Funding: Series C
Headquarters: Seoul, South Korea
Founded: 2017
Glassdoor
Glassdoor: 4.6
WebsiteLinkedInGlassdoor