Superintelligence Lab - Members of Technical Staff

Superintelligence Lab은 다음 영역에서 뛰어난 역량을 보유한 Members of Technical Staff (MTS)​를 채용합니다.

• Move between first-principles research and production-grade engineering.
• Pursue 0-to-1 ideas before they become obvious.
• Care deeply about rigor, speed, quality, and systems that actually work.

Preferred Qualifications

• PhD or postdoctoral experience with strong engineering skills.
• BS or MS degree with production-grade engineering and research experience.
• Domain experts or winners of Olympiads and international competitions.

Areas of Focus

1. Agentic Systems

• Design agentic systems and harnesses for self-improvement.
• Build long-term, episodic, and working memory systems through context engineering and optimization.
• Develop evaluation and verification methods for multi-agent behavior.
• Strong intuition and experience in reinforcement learning for foundation models and orchestrator training.

1. Multimodal Modeling

• Experience in training VLMs, VLAs, image and video retrievers, and world models.
• Develop multimodal agent systems and harnesses.
• Conduct research on visual documents containing tables, charts, and graphs.
• Develop multimodal evaluation methods for visual grounding, retrieval quality, embodied task execution, and complex multimodal tasks.

1. HAI & Evaluation

• Strong experience across both AI and HCI.
• Experience with evaluation problems and incentive design.
• Research tacit knowledge and domain-specific workflows.
• Hands-on experience with A/B testing and product design processes.

1. Inference & Efficiency

• Expertise in neural architectures such as Mixture-of-Experts and large-scale AI systems.
• Work on KV caching, speculative decoding, model routing, parallel decoding, long-context serving, retrieval latency optimization, and tool-call scheduling.
• Design memory-efficient execution systems for agents operating across models, tools, retrievers, and environments.
• Hands-on experience improving LLM serving frameworks such as vLLM and SGLang.