Research Member of Technical Staff (LLM Development)

This role develops large language models, including Sakana Namazu, which serve as foundations for the company’s products and solutions and support its RSI research. The position is a Member of Technical Staff role equivalent to a Research Engineer.

The LLM development team works across pre-training, mid-training, post-training, and evaluation. Responsibilities include improving model architecture, building and refining training data, estimating scaling laws and uncertainty through small-scale experiments, scaling methods to large training runs, and optimizing training efficiency. The role requires ownership of one or more capability domains, technical components, or other areas, with responsibility for leading work in the assigned area and for its effect on the final model and the overall development process.

Work includes setting goals for a capability domain by selecting or creating benchmarks, building goal-aligned data for agentic mid-training, supervised fine-tuning and reinforcement learning, validating and improving data and training methods with small-scale models, and applying training data and recipes to a final large-scale model. The work follows an iterative process of forming hypotheses, implementing them, validating results, and making improvements.

Required qualifications include a master’s degree or equivalent experience, at least two years of practical experience in machine learning or deep learning or equivalent experience, and experience designing, running, and debugging multi-GPU machine learning jobs in a shared compute environment. Candidates must be able to decompose ambiguous problems into testable hypotheses and design experiments that support decisions. The posting also asks for the ability to learn unfamiliar knowledge and skills quickly. Business-level Japanese is listed as a requirement.

Preferred experience includes improving model competitiveness through the collection, filtering, mixing, and quality assessment of training data; hands-on LLM training, especially for LLMs deployed in production, with responsibility for model quality; statistical experiment design and evaluation that produces reproducible findings from noisy or uncertain results; and knowledge of computer architecture and high-performance computing, including FLOPs, memory bandwidth, parallelism, and communication costs. Publications in machine learning, particularly LLMs, and contributions to widely used machine learning open-source software are also preferred.

The role is based in Tokyo, Japan, with a hybrid arrangement combining office attendance and remote work. The CV and cover letter must be submitted in English, and applicants must select and identify the Member of Technical Staff (LLM Development) position in the application form and cover letter. Candidates are encouraged to apply even without every listed qualification, and equivalent experience from other fields is considered.