Japanese-RP-Bench
2026A benchmark for evaluating Japanese LLM role-play performance. Version 2 extends the original benchmark to measure character adherence, long-term stability, resistance to persona replacement, and recovery after failure.
A developer focused professionally on AI agents while also building and operating AI characters as long-term independent projects. Creator of AI Nike-chan and this website.
A web engineer with more than eight years of experience across Java-based business systems and Ruby on Rails web services. Based in Poland and working fully remotely with Japanese companies.
Works end to end across requirements, basic design, technology selection, implementation, infrastructure, launch, and ongoing operation. Current work covers AI agents with conversation design, memory, RAG, and external tool integrations, along with AI-assisted development practices.
Maintains AITuberKit, an open-source project with more than 1,000 GitHub stars, and operates AI Nike-chan through real work and relationships to study how AI characters can participate meaningfully in society.
Rather than stopping at technology commentary, I integrate new models and services into working systems, operate them where people actually interact, and document both failures and improvements. AITuberKit is the platform for building; AI Nike-chan is the long-running case study.
Front, side, and back views for illustration and fan-work reference.
Download high resolution
Maintained and extended a Java-based diagnostic agent management system at a clinical research company.
Worked as a fully remote freelance engineer on contract projects for Japanese companies, primarily developing web services with Ruby on Rails.
Applied expertise gained through independent AI development to professional LLM-based AI agent projects.
Focuses primarily on AI agent development, using AI development tools as part of the standard engineering process.
A benchmark for evaluating Japanese LLM role-play performance. Version 2 extends the original benchmark to measure character adherence, long-term stability, resistance to persona replacement, and recovery after failure.
An AITuber system created using generated videos. By combining pre-prepared videos, it appears as if the character is actually speaking.
An open-source toolkit for building AI character chat and AITuber streaming experiences. It supports a wide range of LLMs, TTS services, and 2D/3D character models, and has grown beyond 1,000 GitHub stars.
A practical AI assistant and public character who grows through real activity and relationships. She is an ongoing case study in memory, relationships, and safe long-term operation.
A website that aggregates AITubers posting on YouTube.
An interactive development support tool that combines Live2D characters with a programming execution environment. Enables intuitive program execution through conversation with anime-style characters.
A project where AI fully automates gameplay and commentary. Automates the entire process from turn-based game strategy to commentary generation.
View the latest three records and materials
I speak and collaborate on AI characters and agents from hands-on experience building and operating them over time. Inquiries from companies, event organizers, developers, and creators are welcome.
We are looking for sponsors to support AI Nikechan's development. All support via FANBOX has the same benefit design, with an open management policy.
Sponsors will have access to a dedicated channel on Discord. However, we aim to share information publicly as much as possible, minimizing exclusive content.
Participation instructions and channel details are provided in a dedicated Discord channel.