downsizing large language models is the future of generative AI
Prodcuts and Solutions
Our Business Record
for Smart City
-
LogisticsAutomation
- Ordering system
-
FactoryRobotics
- Anomaly Detection
-
HospitalOnline Pharmacy
- Prescription Recognition
- Medicine Audit System
-
LifeHotel/ Building
- Cooling and Heating Load Forecasting
- Hotel Dynamic Pricing
-
EnergyEnagy Management
- Solar Plant Anomaly Detection
- Solar Power Generation
Forecast - Power Consumption Forecast
Our Project
Kyoto Pharmacy Project
In the good old days of the Showa era, when you had a stomach ache, it was common to consult a pharmacist at a nearby pharmacy and get medicine. It was a pharmacy that doubled as Health care room.
Currently, pharmacies in front of hospitals and drugstores are the mainstream. The pharmacists are busy with dispensing work, and out of consideration for the patient's waiting time, it seems that they are doing their best to convey the minimum necessary information. We feel like we want to reduce the work of the pharmacist.
In Kyoto, there are many pharmacies that are 40 or 90 years old, and still have the goodness of the town pharmacies. Such pharmacies are in a period of change, with succession problems, financial difficulties due to intensifying competition, the spread of the coronavirus, and the needs of online pharmacies.
As a patient, I want a convenient system and an online purchase system, but on the other hand, I sincerely hope that the goodness of the old-fashioned town pharmacy will be preserved, and I want to reduce the burden on pharmacists with a convenient system. It's been three years since I started the "Kyoto Pharmacy Project".
We are tackling the difficult problem of "blending digitalization and face-to-face goodness".
SmallTrain Open Source ProjectThe project adopted for Kyoto Economic Gardening Program
Our self-developed AI engine "SmallTrain" has been released as an open source project.
SmallTrain is a deep learning framework for AI development, and it's a pre-trained model on all kinds of datasets. You can develop an AI for 1/10th the price in just a few weeks.
- SmallLess data with less developing cost
- TrainTrain of machine learning
We publish the technologies used in our product development.(Compliant with MIT license for commercial use)
SmallTrain AI-OCR
OCR stands for Optical Character Recognition/Reader. A technology that reads handwritten or printed characters with a scanner or digital camera and converts them into digital character codes for computers. AI-OCR is an OCR technology that utilizes AI technology. It is sometimes called ICR (Intelligent Character Recognition) overseas. Machine learning and deep learning are used to learn the results of character corrections, improving the character recognition rate.
In order to use AI-OCR in the office scene, it is necessary to ensure sufficient recognition accuracy and recognition speed. In general, you should use expensive GPU servers. However, if the processing cost (server cost) of 100 yen or more is required to recognize one document, it is not worth it. Our AI-OCR is capable of high-speed processing with a character recognition accuracy of 90% or higher, making it the optimal AI-OCR for actual operation at a unit price of 10 yen or less per page. We are working on research and development to lower the unit price per sheet.
Our AI-OCR is used in pharmaceuticals. It is used for highly accurate prescription recognition AI-OCR and pharmaceutical audit applications that can handle various prescription formats.
We provide AI-OCR API service. Please see our serivce page.
Research and Development of Basic AI Technology
to minimize "the use of large amounts of data"
and "computing power"
SmallTrain does not need "the large amounts of data"
- No need to build environment
- No code (may require a few lines in the input layer)
- Just perform a simple transfer learning
We publish the technologies used in our product development.
(It complies with the MIT license for commercial use)
AI Model Compression Techniques
A deep learning AI model has a hierarchical structure of about 100 layers of neural networks. By reducing the hierarchy, AI can be faster and lighter, delivering results without the GPUs traditionally required.Miniaturized AI can be used on smartphones, which are less powerful devices.
A high-spec server is required to develop and operate a deep learning AI model, and the cost of the server accounts for more than half of the cost of commercialization. Due to the miniaturization of AI, the cost of product services can be lowered, and the price of product services can be lowered.
In order to provide high-quality AI products and services at an affordable price, we researched and developed AI miniaturization technology and succeeded in reducing the size to about 1/10.
NEWS
VIEW MORE-
2023.01.09
exhibition
From January 18th (Wednesday) to 20th (Friday), we will exhibit at "The 4th PHARMACY SOLUTION EXPO OSAKA" at Intex Osaka.
-
2021.07.06
Japan Representative
Our company was selected as a Japanese representative startup in the "Japan National Team Startup Selection Contest" of the "Global Innovation Forum (GIF)".
-
2021.02.26
Award
At the Kansai NBC New Business Awards 2020, we received the New Business Grand Prix and the Kinki Bureau of Economy, Trade, and Industry Director's Award.
-
2021.01.27
Press Release
The source code of the deep learning framework "SmallTrain ver.0.2.1" has been released under the MIT license. This version supports the Jupyter Notebook interface.