One engineer across cloud, security, and machine learning
Your project is handled by an engineer who holds three Google Cloud Professional certifications, has passed Japan's Registered Information Security Specialist Examination, and holds the JDLA E-Certification in deep learning. You can go from design to implementation without splitting the work across vendors.
02
Results backed by numbers
For an AI that predicts CO2 emissions from construction estimates, we cut preprocessing time from 30 minutes to 3 minutes and raised the accuracy of the model that predicts estimate column names from 85% to 96%.
03
We take over undocumented systems
With no administrator and no documentation, we analyzed an Active Directory environment of about 3,000 devices from its settings, redesigned the distribution path, and cut management work by about 60 hours a month while keeping a 100% patch rate.
04
From technical validation to training
We set the implementation direction for a next-generation system through a Rust prototype and technical report, and our training for more than 50 engineers was rated 4.5 out of 5.
Works
Work
Out of respect for confidentiality agreements, client names and specific details are withheld.
Preprocessing time30 min → to3 min
Accuracy of the column-name prediction model85% → to96%
Software development company (AI service for the construction industry)
We cut preprocessing time from 30 minutes to 3 minutes, built retrieval-augmented generation that handles manuals with figures and tables, and set up a training environment where experiments can be tracked.
Python
Pandas
PyTorch
FastAPI
Amazon Bedrock
Amazon SageMaker
About 3,000 Windows devices (1 square = 10 devices)100% patch rate
We analyzed an undocumented Active Directory setup, redesigned the distribution path, and automated progress tracking, cutting about 60 hours of work a month.
Active Directory
Group Policy
WQL
VBA
Participant survey rating4.5 / 5More than 50 engineers trained
Training program hosted by a regional IT industry association