Kotozna Announces Proprietary RAG Architecture 'TocDex RAG,' Revolutionizing Search Structure to Improve Generative AI Chatbot Response Accuracy
Kotozna Inc., a B2B SaaS company providing multilingual communication platforms utilizing generative AI, has released its unique Retrieval-Augmented Generation (RAG) architecture, 'TocDex RAG (Table of Contents + Index RAG).' This new architecture significantly enhances the accuracy, multilingual support, and scalability of generative AI chatbots by employing a two-layer search method combining category and index searches. It addresses traditional RAG challenges in contextual understanding and multilingual search, making it suitable for applications requiring precise terminology and multilingual capabilities.
📋 Article Processing Timeline
- 📰 Published: April 14, 2026 at 20:00
- 🔍 Collected: April 14, 2026 at 11:31
- 🤖 AI Analyzed: April 14, 2026 at 17:37 (6h 6m after Collected)
Kotozna Inc. (Headquarters: Minato-ku, Tokyo; CEO: Genri Goto), a B2B SaaS company that provides multilingual communication platforms utilizing generative AI, has released its proprietary RAG (Retrieval-Augmented Generation) architecture, 'TocDex RAG (Table of Contents + Index RAG),' designed to enhance the response accuracy of generative AI chatbots.
With the implementation of this architecture, the response accuracy, multilingual support, and scalability of generative AI chatbots will be significantly improved.
TocDex RAG adopts a two-layer search method that combines category search (table of contents) and index search (index), and is integrated into Kotozna's enterprise generative AI platform, 'Kotozna TPG.' This new RAG architecture re-examines the search structure itself to address the challenges of contextual understanding and multilingual search that traditional RAG systems faced.
How TocDex RAG Works
In recent years, many companies have been leveraging large language models (LLMs) to implement systems for searching and answering internal knowledge and customer-facing information. However, traditional RAG systems have faced challenges in the accuracy of contextual understanding, support for multilingual queries, and processing large-scale data.
To solve these issues, TocDex RAG employs a two-layer search architecture that combines category search and index search. By organizing and narrowing down information by category while simultaneously performing word-unit searches like an index, it becomes possible to efficiently retrieve more contextually relevant information before the generative AI produces an answer.
Differences Between TocDex RAG and Conventional RAG
Comparing Conventional RAG and TocDex RAG
In conventional RAG, it is common practice to pass text fragments (chunks) obtained through search directly to the LLM to generate answers. However, these chunks are often divided without considering semantic boundaries, which can lead to cases where the LLM does not receive sufficient contextual information.
In contrast, TocDex RAG provides the LLM not only with the retrieved chunks but also with semantically coherent passages that include their surrounding context. This allows the LLM to understand the preceding and following context, enabling the generation of more accurate and natural responses.
Genri Goto, Representative Director and CEO of Kotozna Inc., states:
"The performance of RAG is greatly influenced by the quality of the chunks. Chunks structured in meaningful units enable highly accurate answers, but conventional RAG often does not sufficiently consider this point."
Key Advantages of TocDex RAG
Compared to conventional RAG, TocDex RAG offers the following features:
・Improved response accuracy through enhanced contextual understanding
・Strengthened support for multilingual queries
・Suppressed impact on response speed while cost-effectively processing large-scale data
・No additional maintenance burden required
Features Supporting Scalability and Flexibility
The following features are also implemented for enterprise operation:
・Automatic synchronization of vector databases upon data updates
・Ability to toggle functions ON/OFF per bot
・Customizable keyword search settings
Anticipated Application Areas
TocDex RAG is particularly effective in fields that require precise terminology understanding and multilingual support. For example, in the tourism and hospitality industry, it is necessary to accurately handle highly specific information such as store names, facility names, and product names. It is also suitable for documents containing specialized terminology, such as product manuals and technical documents.
In the future, it is expected to be utilized in various industries as a generative AI foundation supporting corporate knowledge utilization and customer service.
Experience the New TocDex RAG Feature
TocDex RAG is implemented in Kotozna's generative AI platform, 'Kotozna TPG 2.0.' 'Kotozna TPG 2.0' is a no-code platform that allows anyone to easily create generative AI chatbots without specialized knowledge. Free plans are currently available for both corporate and individual users, allowing you to experience the comfortable bot experience provided by TocDex RAG.
Kotozna TPG 2.0 Sign-up (User Registration)
https://www.kotozna.com/ja/tpg
Kotozna TPG 2.0 User Manual
https://prompt-engineering.kotozna.com/manuals/
AI Technology and Characteristics Explained by Kotozna CEO (Over 30 series, approx. 3 minutes each)
https://www.youtube.com/@kotoznaofficial9548
Explaining the 'Concept of Division' that Designers Should Be Aware Of
AI accuracy changes depending on where you divide it.
In this video, Kotozna CEO Genri Goto explains 'meaningful chunks,' a point often overlooked in AI design using RAG and KAG.
https://youtu.be/btxqQlOddEA
About Kotozna Inc.
Official Website: https://www.kotozna.com/ja/about
Representative: Representative Director and CEO Genri Goto
Established: October 2016
Capital: 70,000,000 JPY (as of December 31, 2025)
Location: Hulic JP Akasaka Building 3F, 2-5-8 Akasaka, Minato-ku, Tokyo
Business Activities: Provision of multilingual communication support services utilizing generative AI
Inquiries: sales@kotozna.com
Company Website: https://www.kotozna.com/ja
With the implementation of this architecture, the response accuracy, multilingual support, and scalability of generative AI chatbots will be significantly improved.
TocDex RAG adopts a two-layer search method that combines category search (table of contents) and index search (index), and is integrated into Kotozna's enterprise generative AI platform, 'Kotozna TPG.' This new RAG architecture re-examines the search structure itself to address the challenges of contextual understanding and multilingual search that traditional RAG systems faced.
How TocDex RAG Works
In recent years, many companies have been leveraging large language models (LLMs) to implement systems for searching and answering internal knowledge and customer-facing information. However, traditional RAG systems have faced challenges in the accuracy of contextual understanding, support for multilingual queries, and processing large-scale data.
To solve these issues, TocDex RAG employs a two-layer search architecture that combines category search and index search. By organizing and narrowing down information by category while simultaneously performing word-unit searches like an index, it becomes possible to efficiently retrieve more contextually relevant information before the generative AI produces an answer.
Differences Between TocDex RAG and Conventional RAG
Comparing Conventional RAG and TocDex RAG
In conventional RAG, it is common practice to pass text fragments (chunks) obtained through search directly to the LLM to generate answers. However, these chunks are often divided without considering semantic boundaries, which can lead to cases where the LLM does not receive sufficient contextual information.
In contrast, TocDex RAG provides the LLM not only with the retrieved chunks but also with semantically coherent passages that include their surrounding context. This allows the LLM to understand the preceding and following context, enabling the generation of more accurate and natural responses.
Genri Goto, Representative Director and CEO of Kotozna Inc., states:
"The performance of RAG is greatly influenced by the quality of the chunks. Chunks structured in meaningful units enable highly accurate answers, but conventional RAG often does not sufficiently consider this point."
Key Advantages of TocDex RAG
Compared to conventional RAG, TocDex RAG offers the following features:
・Improved response accuracy through enhanced contextual understanding
・Strengthened support for multilingual queries
・Suppressed impact on response speed while cost-effectively processing large-scale data
・No additional maintenance burden required
Features Supporting Scalability and Flexibility
The following features are also implemented for enterprise operation:
・Automatic synchronization of vector databases upon data updates
・Ability to toggle functions ON/OFF per bot
・Customizable keyword search settings
Anticipated Application Areas
TocDex RAG is particularly effective in fields that require precise terminology understanding and multilingual support. For example, in the tourism and hospitality industry, it is necessary to accurately handle highly specific information such as store names, facility names, and product names. It is also suitable for documents containing specialized terminology, such as product manuals and technical documents.
In the future, it is expected to be utilized in various industries as a generative AI foundation supporting corporate knowledge utilization and customer service.
Experience the New TocDex RAG Feature
TocDex RAG is implemented in Kotozna's generative AI platform, 'Kotozna TPG 2.0.' 'Kotozna TPG 2.0' is a no-code platform that allows anyone to easily create generative AI chatbots without specialized knowledge. Free plans are currently available for both corporate and individual users, allowing you to experience the comfortable bot experience provided by TocDex RAG.
Kotozna TPG 2.0 Sign-up (User Registration)
https://www.kotozna.com/ja/tpg
Kotozna TPG 2.0 User Manual
https://prompt-engineering.kotozna.com/manuals/
AI Technology and Characteristics Explained by Kotozna CEO (Over 30 series, approx. 3 minutes each)
https://www.youtube.com/@kotoznaofficial9548
Explaining the 'Concept of Division' that Designers Should Be Aware Of
AI accuracy changes depending on where you divide it.
In this video, Kotozna CEO Genri Goto explains 'meaningful chunks,' a point often overlooked in AI design using RAG and KAG.
https://youtu.be/btxqQlOddEA
About Kotozna Inc.
Official Website: https://www.kotozna.com/ja/about
Representative: Representative Director and CEO Genri Goto
Established: October 2016
Capital: 70,000,000 JPY (as of December 31, 2025)
Location: Hulic JP Akasaka Building 3F, 2-5-8 Akasaka, Minato-ku, Tokyo
Business Activities: Provision of multilingual communication support services utilizing generative AI
Inquiries: sales@kotozna.com
Company Website: https://www.kotozna.com/ja