LLM Engineer

The LLM Engineer supports the business use of generative AI and large language models. Responsibilities cover the full process from data preparation through model selection, construction, tuning, technical evaluation, and planning for business applications.

Day-to-day work includes structuring, processing, and analyzing data; identifying data needed for model development; designing databases; selecting LLMs; setting up the proprietary Mizuho LLM; implementing and running tuning methods; and analyzing model performance. The role also designs practical tests, detects incorrect answers, establishes logging and access-control rules, and considers suitable applications based on evaluation results. Collaboration with product teams is required to connect LLM capabilities with business-support applications.

The work includes organizing internal knowledge and designing training data for the Mizuho LLM, comparing continued pretraining, fine-tuning, and RAG, and validating combinations of these methods. The role contributes to continuous accuracy improvement, deployment, and evaluation through banking-practice tests covering deposits, lending, foreign exchange, and financial analysis. It may also involve identifying practical issues with business departments, collecting data, designing evaluations, and conducting training trials. Technical validation includes building and operating distributed computing environments and clusters, including on AWS, for large-scale learning.

Named technologies include TypeScript, Python, Next.js, React, Vue.js, FastAPI, Streamlit, Assistants API, Agents SDK, Strands Agents SDK, LangChain, LangGraph, Vertex AI, AWS, S3, ECS, ElastiCache, Terraform, CDK, Bedrock, Bedrock Agentcore, Jest, Pytest, Playwright, GitHub, Teams, Slack, Jira, and Confluence. AI tools listed include ClaudeCode, AWS Kiro, OpenAI Codex, Amazon Q Developer, Devin, Dify, and v0.

Required qualifications include knowledge of machine learning or data science and experience independently advancing development or validation processes involving data preprocessing, training, and evaluation; leading projects through requirements definition, technology selection, and progress management; implementing data processing, validation, evaluation, and improvement using Python or similar tools; or handling technology validation through application planning and implementation with stakeholders. Strong interest in and willingness to learn generative AI technologies, including LLMs, is also required. Preferred experience includes building or additionally training proprietary or large language models, leading machine-learning or generative-AI projects, big-data analysis or AI model development, systems using RAG, search, AI agents, VLM, or OCR, machine-learning platforms such as SageMaker, Azure ML, or Databricks, financial-sector data analysis or AI use, and researching English-language technical documents or papers.

The position is based in Tokyo and uses a hybrid office and remote-work arrangement. The employment contract has no fixed term, with a six-month probationary period and the same salary during probation. Standard working hours are 8:40 to 17:10, including 7.5 hours of scheduled work and a 60-minute break; flextime may apply depending on the department. The position is full-time and permanent. Salary is determined under company rules and is not numerically specified. Benefits and conditions include commuting-cost reimbursement under company rules, lunch support, social insurance, a shareholding plan, babysitter childcare discounts, housing-related programs, asset-building support, leisure support, paid leave, and other special leave. Saturdays, Sundays, public holidays, and specified year-end and New Year holidays are days off.