LightBlue How Leading AI Vendors Handle Essential Training Data for Generative AI
Using the harBest platform to accelerate AI development.
Please tell us about LightBlue!
Certainly! Lightblue's business is centered around two primary areas: "human sensing technology," which primarily involves image processing for human subjects, and "LLM development utilizing natural language processing," which has seen significant growth and adoption in recent years.
Our human sensing technology, particularly image analysis, addresses critical needs across various industries. Our developments are tailored for applications in construction sites and factories, focusing on safety management and understanding human behavior. This technology plays a crucial role in enhancing safety protocols and operational efficiency in dynamic environments.
In the realm of LLM (Large Language Model) development utilizing natural language processing, we are committed to providing cutting-edge solutions that leverage advanced algorithms to address diverse customer challenges. Rather than specializing in a specific niche, our goal is to develop adaptable algorithms that can cater to a wide range of customer needs, ensuring versatility and effectiveness in real-world applications.
Thank you very much. Could you tell us about your career, Mr. Taniguchi?
I graduated from the University of Tokyo and subsequently joined this company after completing my graduate studies. My journey with the company began during my sophomore year when the company's founder, Mr. Sonoda, had just established it. I was eager to gain engineering experience and decided to join the company part-time. Over time, I took on various responsibilities and eventually transitioned into my current role.
Currently, I serve as both a Project Manager (PM) and an engineer. In my capacity as a PM, I oversee project management tasks while also actively participating in project implementation. The balance between PM duties and hands-on engineering work is approximately 60% to 40% or 70% to 30%.
I see. Were you in a laboratory?
In graduate school, I was part of the Aoyama Lab, where I conducted research focused on systems engineering aimed at enhancing productivity. Specifically, my research centered on improving the efficiency of metal press work. This involved exploring methods to optimize metal press processing for greater productivity. During this time, I was simultaneously working on my master's thesis while also engaging in practical projects for the company.
Thank you very much. How many employees do you have now?
We have about 15 employees, and if you include outsourcing, I would say we have about 30 employees.
Please tell us about your first encounter with harBest.
Our company required a significant volume of annotations for our business operations. Initially, we managed our own AWS SageMaker environment for development purposes. However, the cost and resources required to handle annotations internally became a challenge. In search of a more efficient annotation service, we discovered harBest.
What kind of products did you need training data for?
We primarily needed training data for object detection annotations. This involved identifying specific parts within a factory environment and labeling individual heavy machinery for detection, such as identifying heavy machines within tunnels.
May I ask what was the deciding factor for choosing harBest when comparing annotation tools?
The deciding factors for choosing harBest over other annotation tools were primarily the cost of the platform and the range of functionalities it offered. We already had a part-time annotator, so we needed a tool that could complement our existing resources. Additionally, we evaluated the tool based on its support for various tasks like image classification, object detection, and segmentation before making our decision.
Why did you choose harBest over other free annotation tools?
We initially considered using a free annotation tool for smaller projects but found that the actual implementation was challenging due to server costs and the management required for in-house annotators. We ultimately chose harBest because it addressed both challenges effectively.
One of the key appeals of harBest is its user-friendly platform, which allows us to easily place orders based on the desired data format for collection and creation.
Depending on the data format you want to collect or create, harBest makes it easy to place orders from the platform.
Are there any parts of harBest you would like to see improved?
I've noticed that harBest is continuously expanding its capabilities at a rapid pace, which is impressive. The user interface is intuitive and easy to navigate, supported by a straightforward in-house manual that enables new users to get started quickly with minimal guidance.
One improvement I'd suggest is the addition of a filtering function for projects. With the significant increase in project volume, a filtering feature that allows us to sort by time of year, project status, registration date, and other criteria would greatly enhance usability and efficiency.
That will be included soon! Thank you for your feedback. Finally, may I ask about your vision for the future?
Our company's vision is to evolve our services by integrating 'human sensing' with 'LLM' (Large Language Models). Currently, our product development and project assignments are somewhat segmented between image analysis and natural language tasks. However, we aim to enhance the value of our services by integrating the outcomes of image analysis with LLM to automate report generation and insights extraction.
We're focused on expanding our team with talented individuals and increasing project density. Our goal is to undertake larger-scale projects that have a meaningful impact on society. We've received continued interest from clients for such initiatives, and we're excited to gradually take on bigger challenges as we move forward.
Mr. Taniguchi, thank you very much for your valuable time today!
関連事例
-
Streamlining Development After Successful Outsourcing of High-Volume Annotation Work
The Ricoh Company, Ltd
- AI Development (Experienced)
- Annotation
- Data Management/Labeling
- IT/Internet
- Annotation
- Experienced
-
Micro Control Systems: AI “Visualizing” Factories to Enhance Manufacturing
Micro Control Systems
- AI Development (Experienced)
- Annotation
- Data Management/Labeling
- IT/Internet
- Annotation
- Data collection
- Experienced
-
harBest Boosts MiiTel’s AI Speech Recognition
RevComm Inc.
- AI Development (Experienced)
- Annotation
- Data Management/Labeling
- IT/Internet
- Annotation
- Data collection
- Experienced
-
Challenges of Developing “LHTM-2”, a Large-Scale Language Processing Model From Japan.
alt, Inc.
- AI Development (Experienced)
- Annotation
- Data Management/Labeling
- IT/Internet
- Annotation
- Data collection
- Experienced
-
AGRIST Interview: Developing Agriculture Technology to Support the Aging Farmers of Japan
AGRIST
- AI Development (Experienced)
- Annotation
- Data Management/Labeling
- Agriculture
- Annotation
- Data collection
- Data Management
- Experienced