Skip to content

Home

Conversational Generative AI
Fujitsu Kozuchi Conversational Generative AI:
Next-Generation AI Solutions Combining Safety and Precision

Fujitsu's Proprietary Technology for Safely Conducting Business Operations

Fujitsu Kozuchi Conversational Generative AI provides advanced security features, including proprietary hallucination and phishing URL detection, to ensure reliable responses and minimize security risks, enabling safe AI use in business operations.

Let's Experience the Latest Technology Now

Try Fujitsu Kozuchi Conversational Generative AI Now!
Experience a secure environment, proprietary hallucination detection features, and a preview of some functionalities available in the Fujitsu Kozuchi production version, as well as cutting-edge technologies planned for future releases.

The Value Delivered by Fujitsu Kozuchi Conversational Generative AI

  1. Safely and securely utilize your corporate data with generative AI
  2. Add documents as data sources for the LLM
  3. Ensure quality and security when using LLM
    • RAG Technology
      • Document Management Functionality for LLM Data Sources
      • Document Configuration with Retriever
    • AI Trust Technology
      • Hallucination Detection Feature
      • Phishing URL Detection Feature[Preview]
      • Keyword Mask Feature[Preview]

Demo
App Try the Demo Application Here

Overview of Features

Target Industries and Users

Companies aiming to streamline various business operations by integrating AI, such as support tasks like FAQs, decision-making support through data analysis, and development activities including system construction.

Challenges in Utilizing LLMs

LLMs only possess information up to the point of their training, making them unable to handle the latest updates, specialized knowledge, or proprietary information.
While methods like fine-tuning or retraining are available, they are costly and unsuitable for phased implementation.
Additionally, responses may contain "hallucinations," which are plausible-sounding inaccuracies.

Fujitsu's Technological Superiority

  • RAG Technology (Document Referencing Functionality)
    • This technology leverages RAG (Retrieval-Augmented Generation) to enable the use of user-specific information (document data) as a source for responses generated by the LLM. These settings are easy to configure: simply register the documents through the web application interface and link them to the chat functionality.

  • Hallucination Detection Feature

    • This application detects hallucinations in the responses generated. It calculates a metric called the "hallucination score" (ranging from 0 to 100) for each response, allowing users to visualize the reliability of the answers. However, as the hallucination detection feature itself may have limitations or accuracy issues, the final judgment should be made by the user.

  • Phishing URL Detection Feature [Research Portal Edition Preview]

    • When a URL appears in the chat text, the system determines whether it is a phishing URL and provides an immediate assessment. If the URL is identified as a phishing link, an exclamation mark (!) is displayed next to it, preventing users from clicking on the URL (link).
  • Keyword Mask Feature [Research Portal Edition Preview]
    • The system detects and masks keyword stuffing*2 found within the documents registered as input data for the RAG feature. This not only neutralizes malicious keyword stuffing but also disables naturally occurring keyword lists (such as tables of contents or indexes), ensuring that the responses generated by the AI are both safe and accurate.

About the "Commercial Edition" and "Research Portal Edition" of Conversational Generative AI

  • Commercial Edition
    • The application for Conversational Generative AI is integrated into "Fujitsu Data Intelligence Basic" (DI Basic) and is commercially available as part of DI Basic's functionality.
  • Research Portal Edition
    • The "Fujitsu Research Portal" offers an environment where you can explore Fujitsu's latest technologies early, providing free access to the Conversational Generative AI application along with some additional features.

Demo
App Try the Demo Application Here

About the Demo Application

  • Available AI Models
    • GPT-3.5 (powered by Azure OpenAI Service)
    • GPT-4o (powered by Azure OpenAI Service)
    • Claude Instant (powered by Amazon Bedrock)
    • Fugaku-LLM (Only the CPU version is available in the free trial.)
      • Since the processing is executed using a CPU, it may take several tens of seconds to about a minute to receive a response.
      • If you wish to use the GPU version for faster processing, please contact us through the inquiry form on the Fujitsu Kozuchi website.
  • Important Notes
    • Generative AI may produce incorrect or inappropriate content.
    • Please ensure that you verify the output results at your own responsibility and make corrections or adjustments as necessary.
    • There are differences in functionality between this Research Portal Edition and the Commercial Edition (DI Basic).
    • The preview features available in this demo may not be directly reflected in the final commercial product. Thank you for your understanding in advance.
  • List of Used Open Source Software (OSS)
    • OSS list：License information for OSS, etc. used by Conversational Generative AI.

Document

Document Name Description
Application Overview Details about the application overview
Login and Logout Details about logging in and out of the application
Chatting Details about using the chat functionality
Document Management Details about the document management functionality
Hallucination Detection Feature Details about the hallucination detection feature
Phishing URL Detection Feature Details about the phishing URL detection feature
Keyword Mask Feature Details about the keyword masking feature
Troubleshooting and FAQ Details about troubleshooting and FAQs
Exiting the Application Details about exiting the application

Demo
App Try the Demo Application Here