Member of Technical Staff, Inference

We're looking for an inference runtime engineer to push the boundaries of what's possible in LLM and diffusion model serving. Models grow larger. Architectures shift: mixture-of-experts, multimodal, agentic. Every breakthrough demands innovations on the inference engine itself. You'll work at the core of vLLM, optimizing how models execute across diverse hardware and architectures. Your work will directly impact how the world runs AI inference.

• Bachelor's degree or equivalent experience in computer science, engineering, or similar.
• Deep understanding of transformer architectures and their variants.
• Strong programming skills in Python with experience in PyTorch internals.
• Experience with LLM inference systems (vLLM, TensorRT-LLM, SGLang, TGI).
• Ability to read and implement model architectures and inference techniques from research papers.
• Demonstrate the ability to contribute performant and maintainable code and debug in complex ML codebases.

Preferred qualifications

• Deep understanding of KV-cache memory management, prefix caching, and hybrid model serving.
• Familiarity with RL frameworks and algorithms for LLMs.
• Experience with multimodal inference (audio/image/video/text).
• Contributions to open-source ML or system infrastructure projects.

Bonus points if you have

• Implemented core features in vLLM or other inference engine projects.
• Contributed to vLLM integrations (verl, OpenRLHF, Unsloth, LlamaFactory, etc).
• Written widely-shared technical blogs or side projects on vLLM or LLM inference.

• Location: This role is based in Singapore.
• Compensation: Depending on background, skills, and experience, the expected annual salary range for this position is S$200,000 to S$400,000 annually + equity.
• Visa sponsorship: We sponsor visas on a case-by-case basis.
• Benefits: Inferact offers a generous benefits package, including medical, dental, and vision coverage.

Inferact's mission is to grow vLLM as the world's AI inference engine and accelerate AI progress by making inference cheaper and faster. Founded by the creators and core maintainers of vLLM, we sit at the intersection of models and hardware—a position that took years to build.