AI Engineer - LLM Adoption Department (LLMAD)

Job Description:

Business Overview
Rakuten Group, Inc. is a global leader in internet services that empower individuals, communities, businesses, and society. Founded in Tokyo in 1997 as an online marketplace, Rakuten has expanded to offer services in e-commerce, fintech, digital content, and communications to approximately 1.7 billion members around the world. The Rakuten Group has nearly 32,000 employees and operations in 30 countries and regions. For more information visit https://global.rakuten.com/corp/

Department Overview
The AI & Data Division (AIDD) creates powerful, customer-focused search, recommendation, data science, advertising, marketing, price, and inventory optimization solutions to a variety of businesses in the Rakuten group. Our goal is to drive innovation by developing new products and capabilities that deliver significant impact over longer timeframes using AI.
The LLMAD team drives adoption of Rakuten AI and open-source LLMs hosted within our own environment. We're hiring AI Engineers who can be embedded directly inside business units to drive real adoption of in-house large language models in place of third-party alternatives.

Position:

Position Details

As an AI Engineer you will be embedded inside a business unit's engineering team and build with them day to day: implementing agent workflows and integrations, iterating on prompts, running evaluations, and getting features live in production on our in-house models.

You will work on an engagement alongside a senior engineer who sets the overall agent architecture and model direction, and you will take growing ownership of the delivery as you build depth. This is hands-on applied LLM work, close to the product.

<Key Responsibilities>
・Embed inside a business unit's engineering team and implement agent and tool-use workflows for their use case - function calling, multi-step flows, retrieval, guardrails, and fallback handling - writing production code in their codebase.

・Get features from prototype into production and support them there: debugging failures, tightening latency and cost, and fixing what breaks.

・Run task-level evaluations of in-house and open-source models for your use case, and extend and maintain the eval sets and harnesses together with the business unit.

・Design and iterate prompts, output schemas, and tool definitions - versioned and measured against evals rather than tuned by feel - and help narrow quality problems down to the prompt, the retrieval, the model, or the data.

・Support migrations from third-party APIs to our in-house models, and surface the gaps you find to the senior engineer on the engagement and to our model and platform teams.

・Document what worked as reusable recipes, examples, and starter templates for the team's shared library, and help business unit engineers adopt our tools and APIs.
 

Mandatory Qualifications:
・3 to 6 years of professional software engineering experience, including hands-on work building LLM applications.
・Practical agentic engineering experience: you have built agents using tool and function calling or an orchestration framework and have debugged them when they misbehave.
・Working knowledge of prompt engineering: structured outputs, iteration against test cases, and the understanding that prompts need versioning and measurement.
・Strong fundamentals in at least one modern language (e.g., Python, Java, Go), and comfort with APIs, backend services, and testing.
・Curiosity about how models behave - you want to understand why an output was wrong, not just retry it.
・Comfortable working embedded in someone else's engineering team, communicating directly with business stakeholders, and operating with a degree of ambiguity.
 

Desired Qualifications:

・Japanese language ability - a significant plus. Our business unit teams work in Japanese day to day, and being able to work in the language directly makes the embedded model far more effective.
・Retrieval-augmented generation experience.
・Familiarity with evaluation frameworks or LLM observability tooling.
・Exposure to self-hosted inference (e.g., vLLM) or cloud model deployment.

#engineer #applicationsengineer #aianddatadiv