Shisa Model Use Cases Now Available

Shisa.AI has published use cases showcasing how Shisa-related models are being utilized in specialized AI applications.

In an AI agent for nuclear engineering design documentation, “shisa-v2-unphi4-14b” from the Shisa-v2 model family was used as the LLM powering the system. The model supports all three core functions: issue detection, hypothesis-based question generation, and answer analysis.

Key reasons for selecting the model included its Apache 2.0 license, which allows commercial use, and its confirmed ability to run on a development PC with 48 GB of GPU VRAM. Its strong Japanese-language processing capabilities also make it well suited to Japanese design documentation and interaction with experienced Japanese-speaking engineers. The secure, on-premises environment is especially valuable for highly specialized and confidential nuclear engineering documents.

The Shisa-V2 model series was also adopted as the unified foundation of a multilingual multi-agent system. The models serve as the inference engine for agents responsible for translation, knowledge extraction, systematization, and follow-up tasks, supporting functions from translation and summarization to SOP generation.

The system benefits from the model series’ strong translation capabilities among open-source models in the same size class, with a particular focus on Japanese-English and English-Japanese translation. Its inference speed enables low-latency, low-cost real-time responses, while in-house model management allows additional training with confidential technical data in a secure environment.