AI / Machine Learning Engineer

**About the company:** monoya Shibuya-ku, Tokyo monoya is a well-funded startup on a mission to bring Japan’s best craftsmanship to the global stage. We’re reinventing how manufacturing companies grow, starting by digitizing the OEM workflow and enabling high-quality Japanese makers to access international markets with speed and scale, and vice versa. Read more **Responsibilities:** LLM-powered agents – design and deploy multi-modal, tool-using agents that classify inquiries, ask clarifying questions, and draft estimates (RAG pipelines, function-calling, etc.). Vector search & knowledge graphs – build and tune semantic search over Firestore + Weaviate, exploring graph-based representations where useful. Model evaluation – establish repeatable benchmarks, offline/online metrics, and automated regressions so we know when a new prompt or fine-tune is truly better. Prototyping → Production – craft PoCs in notebooks, then convert the winners to clean, tested services running on Cloud Run (Python FastAPI, occasional Go/Rust helpers). Collaboration – pair closely with product, and design to ship features end-to-end. Requirements Have 2–3 yrs building ML or data-intensive systems (industry, or advanced grad work). Write clean Python and are fluent in at least one deep-learning framework (PyTorch preferred; JAX/TensorFlow also welcome). Understand the maths enough to debug when a model or retrieval step misbehaves. Have shipped something with modern LLM tooling—OpenAI, Ollama, vLLM, Hugging Face, LangChain, LiteLLM, etc.—or can show an intense side project. Enjoy explaining trade-offs to non-ML teammates. Like the idea of being the first dedicated ML hire and setting best practices from scratch. Nice to haves While not specifically required, tell us if you have any of the following. Japanese ability Hands-on with Google Cloud AI stack (Vertex AI, TPUs, Cloud Functions, BigQuery). Experience fine-tuning or distilling language models, especially for multilingual tasks (JA-EN). Vector DB ops (Weaviate) and evaluation tooling. Blog posts, or OSS contributions we can read. Familiarity with Go or Rust for high-perf data plumbing. Compensation ¥10,000,000 ~ ¥15,000,000 annually. Stock options available (negotiable) **Requirements:** Have 2–3 yrs building ML or data-intensive systems (industry, or advanced grad work). Write clean Python and are fluent in at least one deep-learning framework (PyTorch preferred; JAX/TensorFlow also welcome). Understand the maths enough to debug when a model or retrieval step misbehaves. Have shipped something with modern LLM tooling—OpenAI, Ollama, vLLM, Hugging Face, LangChain, LiteLLM, etc.—or can show an intense side project. Enjoy explaining trade-offs to non-ML teammates. Like the idea of being the first dedicated ML hire and setting best practices from scratch. **Nice to have:** While not specifically required, tell us if you have any of the following. Japanese ability Hands-on with Google Cloud AI stack (Vertex AI, TPUs, Cloud Functions, BigQuery). Experience fine-tuning or distilling language models, especially for multilingual tasks (JA-EN). Vector DB ops (Weaviate) and evaluation tooling. Blog posts, or OSS contributions we can read. Familiarity with Go or Rust for high-perf data plumbing. **Compensation:** ¥10,000,000 ~ ¥15,000,000 annually. Stock options available (negotiable)