株式会社シンカの機械学習エンジニア (LLM)の求人

【仕事概要】
私たちの会社は多くのコミニュケーションデータを保有しており、それらを活用した新たなAI機能を提供していくにあたり、独自大規模言語モデルをゼロから構築することも視野に入れてユーザーごとに最適なAI機能を提供できる基盤の構築に注力していきます。最先端の研究成果を実用的なスケーラブル・システムへと昇華させ、技術的限界を押し広げることで、社会にインパクトを与えるAI基盤の構築を目指しています。

機械学習エンジニア（LLM）として、アーキテクチャの設計、大規模なデータパイプラインの構築、および大規模分散学習の最適化において中心的な役割を担っていただきます。最先端の技術のキャッチアップをしつつ、スピード感をもってAI基盤の構築・改善を推進してもらいます。当社のAI技術の核となるエンジンの開発に初期から関わっていただけます。

【主な業務内容】
・独自のLLMのアーキテクチャ設計から、大規模クラスタを用いた事前学習の実行・管理。
・分散学習フレームワークを用いた学習効率の最大化とCUDAレベルでの最適化。
・数テラバイト規模のデータセットのフィルタリング、トークナイズ、クリーニングプロセスの構築。
・学習したモデルの性能評価、およびファインチューニングの実施。

■About the Team
Our company possesses a vast and unique repository of communication data. As we move toward delivering next-generation AI features powered by this data, we are focused on building a robust foundation capable of providing optimized AI experiences for every user. This includes exploring the development of proprietary large language models (LLMs) from scratch. Our mission is to translate cutting-edge research into practical, scalable systems, pushing technical boundaries to create an AI infrastructure that delivers profound social impact.

■About the Role
As a Machine Learning Engineer (LLM), you will play a pivotal role in designing architectures, building massive data pipelines, and optimizing large-scale distributed training. You will work closely with researchers to translate theoretical breakthroughs into high-performance model weights. This is an opportunity to contribute to the core engine of our AI capabilities in a fast-paced, mission-driven environment.

■In this role, you will
Innovate and Train: Design and execute the pre-training strategy for our proprietary LLMs using large-scale GPU clusters and our unique communication datasets.

Optimize Performance: Implement and enhance distributed training frameworks (e.g., DeepSpeed, Megatron-LM, FSDP) and optimize kernels for maximum hardware efficiency.

Curate Large-Scale Data: Build and manage robust pipelines for filtering, tokenizing, and cleaning terabyte-scale communication data to ensure high-quality model input.

Evaluate and Align: Develop rigorous evaluation benchmarks and apply alignment techniques such as SFT and RLHF to improve model utility and safety for personalized user experiences.

Collaborate and Lead: Work across cross-functional teams to integrate research advancements into our core product offerings.

■You might thrive in this role if you have
Strong Foundation: A Master’s or PhD in Computer Science, Machine Learning, or a related field (or equivalent practical experience).

Deep Learning Expertise: Demonstrated experience in implementing Transformer-based models and proficiency in deep learning frameworks like PyTorch or JAX.

Distributed Systems Knowledge: Hands-on experience with large-scale distributed training across hundreds of GPUs and an understanding of networking/infrastructure bottlenecks.

Proactive Mindset: The ability to move fast in an environment where problems are often loosely defined, owning challenges from conception to deployment.

・コンピュータサイエンス、機械学習、または関連分野の修士・博士号（または同等の実務経験）。
・Transformerベースのモデルの実装経験、またはPyTorch / TensorFlow等のフレームワークを用いた深層学習モデルの開発経験。
・数十〜数百基のGPUを用いた大規模分散学習の経験、または大規模データ処理の知見。
・日本語にてコミニュケーションが可能なこと

Hybrid Work Policy: This role is based in our [Location] office. We believe in the power of face-to-face collaboration and require at least 3 days of in-person attendance per week.

Bilingual Communication: Professional working proficiency in both Japanese and English is required. You must be able to conduct technical discussions, write documentation, and collaborate effectively with global stakeholders in both languages.