Equipped with a High-Precision AI Chatbot Powered by RAG (Retrieval-Augmented Generation) | BtoB Communication Platform 'Qube'
CSS Inc. has released an AI chatbot function on its BtoB communication platform 'Qube', utilizing RAG to generate highly accurate answers based on internal data.
📋 Article Processing Timeline
- 📰 Published: April 23, 2026 at 22:30
- 🔍 Collected: April 23, 2026 at 14:01
- 🤖 AI Analyzed: April 23, 2026 at 23:00 (8h 58m after Collected)
■ Background and Overview
CSS Inc. (Headquarters: Shinagawa-ku, Tokyo, Representative Director: Manabu Sagawa, hereinafter referred to as 'the Company') has released an AI chatbot function utilizing RAG (Retrieval-Augmented Generation) that highly integrates generative AI with internal data on its BtoB communication platform 'Qube'.
This function allows the AI to directly reference unstructured data such as Qube operating manuals, help, FAQs, and related corporate information, automatically generating secure and highly accurate answers to user questions.
■ Technical Features: A Unique Architecture for High-Precision Search
This AI chatbot implements the following advanced technical processes to prevent search omissions and maximize answer accuracy.
1. Unique '8-Step Query Processing Flow'
The process from data ingestion to answer generation is broken down into eight logical steps.
'8-Step Query Processing Flow'
Document Loading: Specification of a group of documents by the user (e.g., Qube manuals).
YAML Structuring and Saving: Converts content into YAML format and stores it in S3 and DB while preventing information loss.
Precise Chunking: Divides data with a maximum of 500 characters and an overlap of 50 characters.
Embedding: Adopts OpenAI's text-embedding-3-small (1536 dimensions).
Vector DB Storage: Builds a database capable of high-speed and highly accurate similarity searches.
Query Classification: Analyzes and classifies the user's question intent using an LLM (OpenAI's gpt-4o-mini).
Hybrid Search Execution: An advanced search combining vector search and keyword extraction.
Answer Generation: Outputs evidence-based answers based on carefully selected information sources.
2. Pursuit of Accuracy through 'Hybrid Search'
To handle proper nouns and specific terms that simple semantic searches (vector searches) cannot cover, a unique hybrid search logic is implemented.
Vector Search: Uses cosine similarity to extract the top 8 chunks that are contextually close.
Keyword Extraction: Extracts proper nouns from the query and identifies information matching the keywords within the vector search results.
■ Expected Effects
The introduction of the AI chatbot to 'Qube' automates the handling of inquiries that operators previously responded to manually, one by one. It significantly reduces the operational time for 'searching for information, verifying, and drafting answers' that previously occurred with every inquiry, enabling instant responses to users. Even in situations requiring careful manual explanation or investigation due to Qube's diverse functions and complex operating specifications, it achieves a dramatic reduction in support man-hours while simultaneously allowing instant knowledge sharing.
■ Future Outlook
CSS Inc. will continue to accelerate the utilization of AI in 'Qube'. In the future, we plan to implement an 'event search' driven by AI and a 'recommendation function' based on user interests, promoting highly accurate, data-driven matching and further personalization of the user experience.
■ What is Qube?
It is a BtoB communication platform centered on hybrid online and offline interactions. We support 'new encounters and value creation in business.' Through practical tools such as event hosting support and SNS functions, we create direct connections among all business professionals, including executives, engineers, and marketing/sales personnel. We aim to form a next-generation business community that encourages co-creation beyond corporate boundaries and accelerates business growth.
[About CSS Inc.]
Company Name: CSS Inc.
Location: Shinagawa-ku, Tokyo
Representative: Manabu Sagawa, Representative Director
URL: https://www.css-net.co.jp/
Qube Official Website: https://qube-q.com/
[Inquiries Regarding This Matter]
CSS Inc.
Digital Marketing Department
Contact: Shimizu
Email: mail_qube@css-net.co.jp
CSS Inc. (Headquarters: Shinagawa-ku, Tokyo, Representative Director: Manabu Sagawa, hereinafter referred to as 'the Company') has released an AI chatbot function utilizing RAG (Retrieval-Augmented Generation) that highly integrates generative AI with internal data on its BtoB communication platform 'Qube'.
This function allows the AI to directly reference unstructured data such as Qube operating manuals, help, FAQs, and related corporate information, automatically generating secure and highly accurate answers to user questions.
■ Technical Features: A Unique Architecture for High-Precision Search
This AI chatbot implements the following advanced technical processes to prevent search omissions and maximize answer accuracy.
1. Unique '8-Step Query Processing Flow'
The process from data ingestion to answer generation is broken down into eight logical steps.
'8-Step Query Processing Flow'
Document Loading: Specification of a group of documents by the user (e.g., Qube manuals).
YAML Structuring and Saving: Converts content into YAML format and stores it in S3 and DB while preventing information loss.
Precise Chunking: Divides data with a maximum of 500 characters and an overlap of 50 characters.
Embedding: Adopts OpenAI's text-embedding-3-small (1536 dimensions).
Vector DB Storage: Builds a database capable of high-speed and highly accurate similarity searches.
Query Classification: Analyzes and classifies the user's question intent using an LLM (OpenAI's gpt-4o-mini).
Hybrid Search Execution: An advanced search combining vector search and keyword extraction.
Answer Generation: Outputs evidence-based answers based on carefully selected information sources.
2. Pursuit of Accuracy through 'Hybrid Search'
To handle proper nouns and specific terms that simple semantic searches (vector searches) cannot cover, a unique hybrid search logic is implemented.
Vector Search: Uses cosine similarity to extract the top 8 chunks that are contextually close.
Keyword Extraction: Extracts proper nouns from the query and identifies information matching the keywords within the vector search results.
■ Expected Effects
The introduction of the AI chatbot to 'Qube' automates the handling of inquiries that operators previously responded to manually, one by one. It significantly reduces the operational time for 'searching for information, verifying, and drafting answers' that previously occurred with every inquiry, enabling instant responses to users. Even in situations requiring careful manual explanation or investigation due to Qube's diverse functions and complex operating specifications, it achieves a dramatic reduction in support man-hours while simultaneously allowing instant knowledge sharing.
■ Future Outlook
CSS Inc. will continue to accelerate the utilization of AI in 'Qube'. In the future, we plan to implement an 'event search' driven by AI and a 'recommendation function' based on user interests, promoting highly accurate, data-driven matching and further personalization of the user experience.
■ What is Qube?
It is a BtoB communication platform centered on hybrid online and offline interactions. We support 'new encounters and value creation in business.' Through practical tools such as event hosting support and SNS functions, we create direct connections among all business professionals, including executives, engineers, and marketing/sales personnel. We aim to form a next-generation business community that encourages co-creation beyond corporate boundaries and accelerates business growth.
[About CSS Inc.]
Company Name: CSS Inc.
Location: Shinagawa-ku, Tokyo
Representative: Manabu Sagawa, Representative Director
URL: https://www.css-net.co.jp/
Qube Official Website: https://qube-q.com/
[Inquiries Regarding This Matter]
CSS Inc.
Digital Marketing Department
Contact: Shimizu
Email: mail_qube@css-net.co.jp