/ AI & Automation
We separate what AI should do from what people should check
You may be wondering whether AI could make your work more efficient, yet it's hard to see which part of the business to start with. RG begins by taking stock of your operations with you. From there, we separate the tasks AI can take on from the ones that call for human judgment.
Reading documents and faxes, entering data, drafting replies to inquiries, searching internal documents — handing tasks like these to AI can cut the effort they take. At the same time, AI isn't always right. We design on the assumption that a person makes the final check, including screens that make checking easy and the operational flow around them.
We develop and run Flux, our own order automation service, which uses AI to read orders that arrive by fax or phone. If you'd rather not send confidential data outside your company, we can also discuss setups that use a language model running locally. From a small trial through to production, the person who first hears your requirements leads the work.
/ Services
01
We use text recognition (OCR) and AI to read order forms, slips, application forms, and similar documents, and turn them into structured data such as item name, quantity, and date. We flag the places where the reading is uncertain, narrowing down what a person needs to check.
02
We match what was read against your product and customer masters and convert it into a format your existing business system can accept. By handling the hand-off to existing systems via CSV or API, we reduce manual re-keying.
03
We build systems that let AI refer to your internal rules, manuals, and past materials, and answer questions together with the documents the answer is based on. For inquiries from outside the company, we can also propose using the output as draft replies for your staff.
04
We use generative AI (large language models) for tasks such as summarizing and classifying text and drafting standard messages. Rather than adding yet another tool to open, we aim to make it available from the business system screens and chat tools you already use.
05
If you'd rather not send data to outside AI services, we consider setups that use a language model running on your own servers (a local LLM). We have run local language models in-house ourselves, and we draw on that experience to help you balance accuracy and cost.
06
Instead of using AI output as is, we build a step into your workflow where a person checks it before it is finalized. Our design covers how the review screen works and how corrections people make feed into the next round of improvement.
/ Why RG
The first question isn't which AI to use, but which tasks are taking time. We take stock of your operations and separate the tasks AI suits from those it doesn't. If a conventional system would solve the problem better than AI, we'll tell you so plainly.
In Flux, our order automation service, AI reads orders that arrive by fax, phone, or order chat, and a person makes the final check before they become order data. It also handles orders that mix Japanese and Korean. Rather than stopping at a prototype, we face the challenges of keeping AI in daily use in our own operations.
The person who first hears your requirements carries out the validation and development. Where AI fits best depends on knowing the fine details of your operations. We believe avoiding hand-offs matters all the more in AI projects.
We can hold meetings in Japanese, English, or Korean. For work that involves documents or inquiries in several languages, we can also help you decide how to read and review each language.
/ Process
STEP 01
We ask which tasks take how much time, and who does them. Looking at real examples of your documents and inquiries, we narrow down the tasks AI could take on. Consultations are free.
STEP 02
We focus on a single operation and build a prototype with real data. Rather than rolling out company-wide right away, we run a proof of concept (PoC) on a small scale to see whether it is likely to pay off before full development.
STEP 03
We assess the prototype's results not only on how accurately it reads or answers, but also on how much effort human review takes. Then we decide with you whether to move to production, change the approach, or hold off.
STEP 04
We set up integration with your existing business systems, review screens, per-user permissions, and so on, to make it ready for daily use. This is also when we decide and implement where data is stored and what may be sent to outside AI services.
STEP 05
Once in use, unexpected document formats and questions appear. Under a monthly maintenance and operations contract, we keep checking accuracy, refining the instructions given to the AI, and adapting to changes in your business.
/ Pricing
Cost depends on the type and volume of documents and data involved, the accuracy you need, whether integration with existing systems is required, and the AI setup used. Starting with a small trial lets you see the results before deciding whether to invest in production. Estimates are free.
All prices exclude consumption tax.
/ Examples
An order automation service for wholesalers and distributors. AI reads orders that arrive by fax, phone, or order chat, and matches them against the product master to create order data. Rather than leaving everything to AI, it is designed so that a person makes the final check. It also handles orders that mix Japanese and Korean. A pilot deployment with a food wholesaler began in October 2026.
In 2026, we took part in developing a SaaS company's AI product for a specialized field. We joined the development team in an engineering role and contributed to building the product.
In 2025, we took on an AI chatbot for restaurants, from design through operations. It uses RAG: the generative AI answers by referring to information prepared in advance.
In 2025, we built three kinds of AI into an illustration-sharing social app: AI that analyzes the content of posts, AI aimed at copyright protection, and AI that automatically checks comments.
/ FAQ
Yes. It's enough to start with "this task is taking a lot of time." Once we've heard about your operations, we'll tell you whether AI is a good fit or whether a conventional system would be better. Consultations are free.
It varies greatly with the format and condition of the documents and the content involved, so we cannot promise figures in advance. We start with a small trial on your real documents and data, confirm the accuracy, and then you decide whether to move to production. We build in a step where a person checks and confirms the results, on the assumption that AI will make mistakes.
Order forms that arrive by fax are among the documents our own product, Flux, reads. That said, accuracy varies with faded text and handwriting habits, including on handwritten documents. We test with your actual documents and then decide how much to leave to AI and what a person should check.
Using a language model that runs locally makes it possible to keep data from being sent to outside AI services. We have run local language models in-house ourselves. Accuracy, cost, and operational effort differ from outside services, though, so we propose a setup that fits your data and goals.
Before work begins, we agree with you on the scope of data we receive, where it is stored, and what may be passed to the AI. If you would like a non-disclosure agreement or similar, please let us know.
It depends on the volume of data, the accuracy you need, and the systems to be integrated, so there's no single answer. We first present the scope and cost of a small trial, then plan production based on the results. Using outside AI services also incurs usage-based fees. Estimates are free.
We propose integration by exchanging data via CSV files or APIs. Flux, too, can pass confirmed order data to existing systems as CSV. What is possible depends on the specifications of the system you're connecting to, so we check this when we talk with you.
Changes such as a new document format or more products can make AI results less accurate. Under a monthly maintenance and operations contract, we keep checking accuracy and refining the instructions given to the AI and the data it refers to.
/ Contact
You don't need a finished spec. Tell us about the problem you're facing, and we'll propose an approach and a rough estimate. Consultations and estimates are free, and we usually reply within one business day.
TEL +81-3-5830-7587 (Weekdays 10:00–19:00 JST)
MAIL info@raregleam.com