Are we "utilizing" log data? Capy Inc.'s initiative to advance analysis using LLM featured in ITmedia
Capy Inc. was featured in ITmedia Enterprise for developing an advanced log data analysis platform using Google's LLM 'Gemini', BigQuery, and Shodan in collaboration with grasys Inc.
📋 Article Processing Timeline
- 📰 Published: April 8, 2026 at 19:30
- 🔍 Collected: April 8, 2026 at 11:00
- 🤖 AI Analyzed: April 20, 2026 at 19:53 (296h 53m after Collected)
Capy Inc. (Headquarters: Chiyoda-ku, Tokyo; President and CEO: Mitsuo Okada) announces that our initiatives regarding the utilization of log data have been introduced as a case study in ITmedia Enterprise.
This article features a log data analysis platform utilizing an LLM (Large Language Model), built jointly with grasys Inc. It is introduced as an initiative to decipher log data, which was previously not fully utilized, including its background and context, and to leverage it for decision-making.
▼ Click here for the published article
https://www.itmedia.co.jp/enterprise/articles/2604/08/news003.html
■ Challenges in utilizing log data
At Capy, in the operation of our unauthorized login prevention service "Capy CAPTCHA," we accumulate log data daily, including source IP addresses and regional information. While this log data contains important information for grasping signs and trends of attacks, there were challenges in directly utilizing the accumulated data for decision-making due to reasons such as:
- The sheer volume of data
- The necessity of specialized knowledge
- The tendency for analysis work to become dependent on individual skills
■ Construction of a log data analysis platform utilizing LLM
In this initiative, in collaboration with grasys Inc., we constructed a mechanism that combines log data accumulated in a data platform (BigQuery) built on Google Cloud with information from the IP intelligence service "Shodan," and performs analysis using Google's LLM "Gemini."
Rather than processing the log data as is, by extracting and organizing necessary information according to the analysis purpose before inputting it into the LLM, we are aiming to improve analysis accuracy and optimize costs.
Furthermore, it is designed to reproduce the process of deciphering the background and context of log data and IP information based on the expertise of security professionals.
■ Streamlining analysis work and supporting decision-making
With the introduction of this system, log data analysis, which conventionally required a certain amount of time, has been streamlined, and the automation of report creation has been realized.
The generated reports organize the location of IP addresses, information on the infrastructure being used, related risk factors, etc., and are provided in a format where the overview is easy to grasp even without specialized knowledge.
As a result, this leads to:
- Facilitation of discussions within the security team
- Acceleration of explanations to and responses for customers
- Increased efficiency in grasping attack trends
In addition, it supports decision-making such as judgments based on log data and considerations of response policies.
■ Towards security that catches signs and anticipates
At Capy...
This article features a log data analysis platform utilizing an LLM (Large Language Model), built jointly with grasys Inc. It is introduced as an initiative to decipher log data, which was previously not fully utilized, including its background and context, and to leverage it for decision-making.
▼ Click here for the published article
https://www.itmedia.co.jp/enterprise/articles/2604/08/news003.html
■ Challenges in utilizing log data
At Capy, in the operation of our unauthorized login prevention service "Capy CAPTCHA," we accumulate log data daily, including source IP addresses and regional information. While this log data contains important information for grasping signs and trends of attacks, there were challenges in directly utilizing the accumulated data for decision-making due to reasons such as:
- The sheer volume of data
- The necessity of specialized knowledge
- The tendency for analysis work to become dependent on individual skills
■ Construction of a log data analysis platform utilizing LLM
In this initiative, in collaboration with grasys Inc., we constructed a mechanism that combines log data accumulated in a data platform (BigQuery) built on Google Cloud with information from the IP intelligence service "Shodan," and performs analysis using Google's LLM "Gemini."
Rather than processing the log data as is, by extracting and organizing necessary information according to the analysis purpose before inputting it into the LLM, we are aiming to improve analysis accuracy and optimize costs.
Furthermore, it is designed to reproduce the process of deciphering the background and context of log data and IP information based on the expertise of security professionals.
■ Streamlining analysis work and supporting decision-making
With the introduction of this system, log data analysis, which conventionally required a certain amount of time, has been streamlined, and the automation of report creation has been realized.
The generated reports organize the location of IP addresses, information on the infrastructure being used, related risk factors, etc., and are provided in a format where the overview is easy to grasp even without specialized knowledge.
As a result, this leads to:
- Facilitation of discussions within the security team
- Acceleration of explanations to and responses for customers
- Increased efficiency in grasping attack trends
In addition, it supports decision-making such as judgments based on log data and considerations of response policies.
■ Towards security that catches signs and anticipates
At Capy...