Machine Learning Engineer (Deep Learning), Credit ML - Merpay

Work Responsibilities

• Design, development, and improvement of core large-scale deep learning models for next-generation credit modeling:

  • Model Architecture Design: Designing deep learning model architectures based on the specific challenges and data characteristics of the credit domain.
  • Training Strategy: Designing training tasks and objective functions; formulating and executing scaling policies considering data, model architecture, and compute budgets.
  • Training Infrastructure: Designing and implementing GPU-based large-scale and distributed training; optimizing and stabilizing model training processes.
  • Evaluation Design: Designing evaluations that encompass not only offline metrics but also key business KPIs, risk metrics, stability, fairness, and explainability critical to credit operations.

• Production Deployment & Operations for Large-Scale Deep Learning Models:

  • Designing training and inference pipelines, optimizing inference efficiency, and integrating models into production systems.
  • Designing operations for monitoring, retraining, and rollbacks, while building and maintaining robust MLOps practices.

• Technical Decision-Making: Determining modeling policies and trade-offs (e.g., model scale/performance vs. cost, inference latency, interpretability, and coexistence with legacy models).

• Research & Application: Catching up with the latest research trends in fintech with deep learning and evaluating their application to in-house challenges.

• Model operations & Continuous improvements: Operating and continuously improving existing credit models.

• Leading consensus-building and collaboration involving stakeholders in product, business, legal, and compliance.

• Mentorship & Knowledge Sharing: Providing technical mentorship in deep learning to team members and systematizing knowledge.

---

Unique Challenges

• Build Next-Gen Models from Scratch: Create next-generation deep learning models in the financial domain from the ground up, utilizing the massive and diverse dataset from Mercari—Japan’s largest marketplace app with 24 million MAU.
• Pioneer Advanced Deep Learning in Credit: Challenge yourself with cutting-edge initiatives in Japan that apply state-of-the-art deep learning and large-scale training expertise directly to credit scoring and risk evaluation.
• High-Stakes Production Implementation: Go beyond R&D to deliver high-stakes, socially responsible production systems, tackling the challenge of creating deep learning models that balance accuracy, stability, explainability, and operational quality.
• Social Impact & Financial Inclusion: Embrace "Go Bold" challenges under real-world constraints—such as complex data interpretation, governance, regulations, and risk management—to generate both business value and a broader social impact through financial inclusion.
• High Ownership & Career Growth: Join a project in its active development phase, where you will hold significant autonomy in deciding model architectures and training strategies, with opportunities to expand your role into project leadership.

---

Qualifications

Required Experience/Skills

• Someone who can align with the mission and values of the Mercari Group and its companies.
• 3+ years of experience in deep learning model development using deep learning frameworks such as PyTorch.
• 1+ year of hands-on experience in pre-training 1B+ parameter Transformer-based models, LLMs, or foundation models from scratch.
• Proven experience independently leading model training, experiment design, and result analysis.
• 1+ year of practical MLOps experience in deploying and operating deep learning models in production environments (or equivalent experience).
• Strong communication skills to collaborate smoothly with team members and cross-functional stakeholders.

Preferred Experience/Skills

• Experience leading the design of data architectures, model architectures, objective functions, training recipes, and compute budgets for pre-training.
• Experience adapting and evaluating pre-trained models for downstream tasks (e.g., fine-tuning, transfer learning).
• Experience with training acceleration, optimization, and tuning (e.g., distributed training).
• Experience optimizing and accelerating inference (e.g., model compression techniques like knowledge distillation, quantization, pruning, or KV caching).
• Experience in tabular data analysis and modeling (deep learning or non-deep learning).
• Experience in data analysis and system development in cloud environments such as AWS or Google Cloud.
• Experience as a team/project lead, as well as experience mentoring junior engineers.
• Experience working in financial institutions or Fintech companies.

Language

• Japanese: Proficient (CEFR - C1)
• English: Independent (CEFR - B2) Preferred

---

Recruiting Cycle at Mercari Group

1. Application screening
2. Skill assessment: For engineering positions, you will be asked to complete a take-home skill assessment. For non-engineering positions, you may be asked to complete an assessment depending on the position. (The timing of the assessment may coincide with the interview process.)
3. Interview: The number of interviews may vary depending on the position.
4. Reference check: We will ask for online references around the timing of the final interview.
5. Offer: Offers will be determined carefully in consideration of the final interview and the reference check.

---

Equal Opportunity Hiring

Here at Mercari, we work to realize a world in which no one’s potential is limited by their background and everyone has the opportunity to freely create value. We also firmly believe that a mindset of Inclusion & Diversity is essential for us to achieve our mission.

This, of course, extends to our hiring practices as well. Mercari is committed to eliminating discrimination based on age, gender, sexual orientation, race, religion, physical disability, and other such factors so that anyone who shares our mission and values can join us, regardless of their background. For more details, please read our I&D statement.

---