【AI/Data Research Engineer】JLPT N2 | Flextime & Hybrid Work

★AI/Data Research Engineer | AI Development Platform Provider

• Business Level Japanese Required

◆ Flextime & Hybrid Work

◆ Start-up Company

◆ Annual salary: 6 million yen - 15 million yen

-------------【About the company】-------------

Adopting a "Data-Centric AI Development" approach, the company specializes in the collection and creation of "high-quality data," which directly dictates the accuracy and safety of AI models. While it is often said that data preparation accounts for 80% of the AI development process, the company addresses this challenge through its proprietary platform.

1. In-house Product: AI Data SaaS

The company operates a platform where crowdworkers from around the world participate in annotation (creation of training data) by utilizing a mechanism similar to "poi-katsu" (reward-earning) apps. By leveraging proprietary quality-control algorithms, the company achieves low-cost and rapid data collection.

2. Data Collection and Annotation Services

The company handles a wide range of projects, from "0 to 1" new development to large-scale operational tasks based on well-defined requirements. It addresses diverse orders in cutting-edge technical fields—such as Large Language Models (LLM), Retrieval-Augmented Generation (RAG), voice recognition, and multi-modal AI—through its flexible and advanced development capabilities.

3. Provision of High-Quality Data Sets

The company provides "rights-cleared" data sets that are essential for AI training, ensuring both quality and legal compliance for its clients.

■Strengths and Track Record

The company’s client base primarily consists of enterprise-level organizations. It supports numerous top players across various industries, notably including the provision of Japanese data for Meta’s Large Language Models (LLM). Rather than being just an AI tool provider, the company has garnered significant market attention as a key player that controls "data"—the very essence of AI.

-------------【 Job Description】-------------

■Recruitment Background

The company provides products that facilitate the generation, preparation, and evaluation processes of training data essential for AI model development.

In recent years, the industry has entered a phase where the "data design and evaluation loop" determines a product's competitiveness even more than the model itself. This shift is driven by several concurrent trends:

・The increasing sophistication of models such as LLMs, VLMs, and image/video models.

・A rapid escalation in data volume and quality requirements for production-level deployment.

・Diverse and customer-specific "data requirements" that directly dictate model performance.

Currently, the company faces challenges where data design and evaluation improvements remain dependent on specific individuals. While the bottleneck for model improvement lies on the data side, these processes have yet to be fully systematized. To address this, the company is looking to welcome engineers as core R&D members who can enhance AI model performance specifically from a "data perspective."

■Role

This is not a position for building models. Instead, this is an engineering role focused on ensuring models "function effectively in production" through strategic data design.

Core Responsibilities

・Training Data Design: Designing training data for LLMs, VLMs, and image/video models.

・Process Design & Optimization: Establishing and improving data preprocessing and annotation guidelines.

・Model Evaluation & Analysis: Designing model evaluation metrics and analyzing evaluation results.

・Feedback Loop Construction: Building and maintaining a continuous "Training → Evaluation → Data Improvement" feedback loop.

・Cross-functional Collaboration: Working closely with model development and product teams.

The essence of this role is to analyze from a data perspective—identifying "why accuracy is lacking" and "which data needs to be modified and how"—and then executing those improvements.

■Core Focus (Primary Expectations)

・Data Design, Preprocessing, and Quality Control for AI Models: Leading the foundational data strategy that powers model performance.

・Design of Model Evaluation and Data Improvement Loops: Building the systems that allow for iterative performance gains.

・Data Processing and Analysis using Python: Utilizing Python as the primary tool for manipulating and interpreting large-scale datasets.

■Examples of Data Types/Domains

・Text, Image, and Video Data for LLMs and VLMs.

・Datasets for Object Detection and Image Recognition models.

・Annotation Data and Metadata management.

*The primary responsibility is not model weight design or algorithmic research.

The core of this role lies in the "Practical Operational Design" of the relationship between Data and Model Performance.

■Discretion and Decision-Making Authority

In this role, you will take proactive ownership of data-related decisions as an engineer, with authority over:

・Defining Data Design Strategies and Quality Standards: Establishing the fundamental blueprints and benchmarks for training data.

・Designing Model Evaluation Metrics and Methodologies: Determining how to measure success and ensure model reliability.

・Decision-Making for Annotation Improvements and Redesign: Taking the lead on when and how to overhaul data labeling processes to boost performance.

・Leading Data Strategy for Model Optimization: Orchestrating the overarching data roadmap specifically aimed at enhancing model capabilities.

■What you'll gain in this role

Hands-on experience in data design that impacts AI model performance

A perspective on fully leveraging cutting-edge models like LLM/VLM

Insights from operationalizing the cycle: training → evaluation → improvement

-------------【 Requirements】-------------

【Required】

While it is not necessary to meet all of the following criteria, the company places the highest priority on proactive experience in "independently conceptualizing and improving" training data and evaluation designs.

・Practical experience working with machine learning models (classification, detection, generative models, etc.) and contributing to the design and improvement of training data to enhance model performance (research or product roles acceptable)

・Experience independently designing and implementing preprocessing, data transformation, and validation for training and evaluation datasets using Python.

・Experience analyzing model evaluation results to determine how data changes could improve performance, and then implementing those changes (e.g., re-annotation, data augmentation, distribution adjustment, label definition review).

【Preferred Requirements】

・Experience in designing data structure, granularity, and labeling schemes for LLM/VLM/image/video models based on task characteristics

・Experience in designing annotation guidelines, establishing quality standards, conducting reviews, and operating improvement cycles

・Experience interpreting evaluation results using metrics like accuracy, recall, F1 score, BLEU, and performing bottleneck analysis

・Experience building data pipelines and RLHF (Reinforcement Learning from Human Feedback) infrastructure with a focus on the learning → evaluation → improvement cycle

・Interested in development that values data design as well as models, based on concepts like Data-centric AI / MLOps

・Track record of research community outputs, such as paper presentations at conferences or peer-reviewed publications

・Experience developing with generative AI (LLMs/AI agents, etc.), with knowledge/implementation experience (or strong interest) in at least one of the following: agent design, tool integration (e.g., function calling), RAG/search, multimodal document understanding (VLM/OCR), safe operation (fact-checking/guardrails), or continuous evaluation via task-specific benchmarks

・Experience with learning methodologies including reinforcement learning, the DS process from problem definition to analysis, implementation, and verification, and the utilization/operation of AI-assisted coding tools

・Experience in data creation within the Robotics/Physical AI domain (e.g., collecting data for IL, generating data for VLA learning, designing evaluations)

【Ideal Applicants】

・Finding more value in “enhancing performance through data and evaluation” than in models alone

・Interested in working on AI products deployed in real-world operations, not just research

・Eager to lead data design while collaborating with model developers

・Willing to take responsibility as the “backend backbone” of AI products

*This role is not suitable for those solely focused on pure algorithmic research.

--------------------------------------------------

◾️Holidays & Leave

・Annual Paid Leave

・Saturdays, Sundays, and National Holidays

・Summer Vacation

・Year-End and New Year Holidays

・Special Leave for Family Events

・Maternity Leave

・Childcare Leave

・Menstrual Leave

◾️Employee Benefits

・Full Social Insurance Coverage

・Full transportation expense reimbursement

・Health checkups

・Influenza vaccinations

・Book purchase expense coverage

・Conference participation fee coverage

・In-house study sessions

・Company meal expense support

・1-on-1 meetings

・Free drinks

・Monitor loan

・Vending machines available

-------------------------------------------------------------------------