＃AskDona

What is RAG?
‍How does AskDona's RAG work.

What is RAG?
RAG (Retrieval-Augmented Generation) is a technology that combines search for external information with text generation using Large Language Models  (LLM).
‍
The basic RAG process
‍
Search phase (Retrieval-Augmented): Take questions from users and search databases and documents for relevant information. At this stage, highly relevant information is selected and prepared for generation.
‍
Generation phase (Generation): A prompt combining “selected information” + “user questions” + “advance instructions” is entered into the LLM and text generation results are returned.

For a more detailed explanation of RAG, check out our blog “What is RAG? Explain the mechanism of RAG and prior knowledge before implementation”

What is ASKDona's RAG?
AskDona's RAG not only allows you to search external data, but it is also possible to link custom tools that can perform tasks. By simultaneously sending external data and custom tools to LLM, it is possible to respond appropriately to user questions or execute tasks more flexibly. AskDona has achieved even more advanced information retrieval and generation by integrating custom tools in addition to RAG technology.
‍
How does AskDona work

Retrieval-Augmented phase : Questions from users are received, and related information organized using OCR technology is advanced searched from databases and documents. At this stage, we use AskDona's unique advanced search to select relevant information for responses and prepare for generation.
‍
Generation phase : Prompts combining “selected information” + “user questions” + “advance instructions” + “custom tools” are input into a LLM, and results of answers executed by text generation or custom tools are returned.

Please see this blog for details on task-oriented technology “What is FunctionCalling in large-scale language models?”
Strengths of AskDona's RAG
RAG has become popular as a common technology, but there are several hurdles until it is actually introduced. AskDona has built a RAG system that is easy for businesses to implement.
‍
Data Formatting Technology Using OCR

The external RAG data is converted into a vector which is an array of numerical values and stored in a state where advanced search is possible. At this time, if the original data to be read is not structured, it becomes noise when vectorized, and vector search accuracy decreases, and as a result, response accuracy decreases. By adopting OCR technology, AskDona can accurately read data from complex materials such as diagrams and tables included in manual data.
‍
Advanced search technology directly linked to answer accuracy

Defining search conditions, rules, etc. is important in order to select highly relevant information in response to user questions. AskDona has adopted a mechanism to drastically improve response accuracy from about 60% (general RAG response accuracy) to about 90% by studying the metadata, order, and number of searches applied to information. A patent has now been filed for a vector search mechanism.
‍
Task execution tool collaboration technology

RAG not only provides appropriate answers, but can also be executed in response to user instructions. AskDona can split and process tasks in response to user instructions and execute tasks appropriately. Task-oriented generative AI assistants are an area that is expected in the future, and it is expected that the work will be completed simply by having the generative AI assistant execute everything by entering forms and linking with the system.

Multilingual support technology

Multilingual support is a basic feature of LLM (large-scale language model), but not all generative AI chatbots that are commonly introduced are multilingual. AskDona has always adopted the latest LLM model and aims to continue to be a generative AI assistant capable of multilingual support, and recognizes that multilingual support is an important factor in supporting business growth. We currently support over 85 languages.

AskDona use cases
AskDona can be used regardless of industry or business type. Many customers are aware of the new way of doing business for the first time after introducing AskDona. Why don't you leave the work you're currently doing to AskDona?

Internal knowledge search

At AskDona, employees can handle the work of inquiring about general affairs, accounting, and human resources. By having AskDona learn internal regulations and sharepoint information, it is possible to answer appropriate questions or share the location of information. By linking AskDona with Slack, ChatWork, and Microsoft Teams, you can have a seamless conversation with AskDona.

Responding to external inquiries
‍

AskDona is also suitable for customer aftercare and support. By having AskDona learn manuals, user guides, and past inquiries and FAQs, it is possible to handle multiple people simultaneously. Also, by linking the inquiry form as a custom tool, it is possible to execute transfers, reservation inquiries, etc. to manned support.

Customer support via SMS
‍
There was an issue where it was difficult to take action from customers when contacting customers by phone number, such as property previews or reminders. AskDona enables two-way communication via SMS. AskDona can automatically respond to questions via SMS, such as location, date and consultation matters.

Customer engagement on social media



Companies that promote engagement with customers on SNS such as LINE official accounts can not only unilaterally distribute campaign and questionnaire information, but also enable two-way communication where users can learn about services while having an intrinsic conversation with AskDona. AskDona can more clearly understand each customer's individual preferences and preferences, and can provide the right service at the right time.
＃AskDona

Everyday tasks with AI assistants

Contact

#AskDona

TOPTerms of useprivacy policy運営会社
Article blog
Copyright © GFLOPS Co., Ltd. All Rights Reserved.