Fujitsu Kozuchi Enterprise AI Factory
Sovereign AI platform for autonomously operating enterprise‑exclusive AI
A Dedicated AI Platform for Safely Developing and Sustaining Task‑Optimized AI
A dedicated, sovereign AI platform that enables enterprises to drive the full generative‑AI lifecycle—development, operation, and incremental learning—within a secure, closed environment. It integrates the high‑precision LLM Takane, in‑house fine‑tuning to refine domain‑specific models, quantization for model size reduction, Generative AI Trust technologies, and, among others, low‑code agent‑development tools. By unifying these capabilities, the platform lowers adoption barriers and supports scalable, long‑term AI utilization across the organization.
| Fujitsu Kozuchi Enterprise AI Factory features | Overview |
|---|---|
| Infrastructure for dedicated environments | Private AI Platform on PRIMERGY / Private GPT enables the use of generative AI in a dedicated environment where data never leaves the organization. Flexible deployment options and comprehensive support from implementation to advanced operations are provided. |
| Generative AI trust technologies | Scanners and guardrails cover over 9,000 vulnerabilities, detecting attacks and unsafe outputs while suggesting applicable rules to ensure high security and reliability. |
| High-precision generative AI models | Fujitsu’s enterprise-grade LLM combines strong Japanese language performance with image analysis capabilities. It supports on-premises deployment, ensures high data confidentiality, and allows customization for specific business needs. |
| In-house Fine-tuning | Enables organizations to train and continuously improve generative AI models using their own data. Accuracy can be optimized for specific business processes and industries within a dedicated environment. |
| Model quantization | Reduces model size while maintaining accuracy. By significantly lowering memory usage and improving computational efficiency, it enhances usability and reduces the cost of AI operations. |
| Efficiency technologies for AI agent development | A low-code/no-code AI agent platform enables rapid, business-led development. Through MCP and agent collaboration, it integrates with existing systems to support advanced AI utilization. |
Solutions
Leveraging a deep understanding of industry-specific terminology and specialized knowledge,these enterprise LLMs generate responses and content in polite Japanese through RAG (Retrieval-Augmented Generation) and fine-tuning.This approach enables the creation of highly accurate information, thereby achieving significant efficiency gains across a wide range of industry- and business domain-specific areas where AI capabilities were previously limited.
Fujitsu Kozuchi Enterprise AI Factory Introduction Materials
Fujitsu Kozuchi Enterprise AI Factory is a dedicated AI platform (with on-premises support) that enables enterprises to use generative AI securely and with high accuracy. It functions as an “enterprise AI factory,” allowing companies to produce and operate business-specific AI models and AI agents in-house without exposing confidential data externally.
Recommended information
GenAI solution with data sovereignty-PrivateGPT
Private GPT is an on‑premise generative AI solution that enables enterprises to securely access and interact with their own data. All processing is performed locally, ensuring data sovereignty and compliance. The turnkey solution includes customized AI models, hardware, software, setup, and support for trusted enterprise AI adoption.
Generative AI trust technologies-LLM Vulnerability Scanner & Guardrails
This technology comprehensively evaluates diverse attack methods against generative AI, automatically detecting and assessing vulnerabilities. Based on the results, it generates and applies optimal defense rules to large language models, enabling secure, reliable, and trustworthy operation of generative AI systems in enterprise environments.
This technology drastically compresses LLM weight data to reduce memory usage and computational load. By significantly improving efficiency, models that previously required large GPU clusters can now run on much smaller GPU configurations, enabling cost‑effective and flexible deployment of large language models.
