Stories of companies that succeeded in transforming AI with Letsur

Early prediction of spinal collapse in patients with osteoporotic spine fractures is an important challenge in the field of neurosurgery. Previously, prognoses were predicted using only clinical numerical data, but there were limitations and the development of an AI model to analyze MRI images was necessary, but it was difficult due to the lack of AI specialists who could handle related image data.

In addition to existing clinical figures, we have developed a new AI model for predicting prognosis based on MRI images. We collaborated on the entire process from data planning and collection to model design, and iteratively upgraded the model to improve accuracy through close feedback from medical staff and AI experts after the first model.

The prognosis of patients can now be predicted more accurately based on 3D medical images, and early treatment and response are possible based on prediction results. Fast and accurate predictions are providing practical assistance to medical staff in their decisions.

In a situation where inquiries related to work manuals continued to occur, a small number of personnel were directly responding to more than 100 inquiries every day. Telephone inquiries were frequent due to various information channels that required finding relevant documents directly from collaboration tools or company drives, which hampered overall work efficiency.

We have introduced an in-house AI chatbot that automates questions and answers based on internal regulations and guideline data. Users can easily search for desired information using keywords, and repeated queries are automatically answered. Chatbots can be used on both PC and mobile, and are naturally linked to existing collaboration tools such as MS Teams.

More than 3,000 simple inquiries repeated each month have been automated, greatly reducing the burden on practitioners and reducing the demand for telephone responses. Based on real-time responses and easy accessibility, company-wide information retrieval speed and employee satisfaction were improved.

In order to plan a new product, it was necessary to review all existing products, and sketching work was expensive and labor-intensive because they relied entirely on designers. As a result, the planning-design process was less efficient, and the time required to prepare for production was excessive.

An AI assistant that provides real-time feedback has been introduced, and planning data is automatically analyzed so that direction can be determined quickly. Sketch was automated using image generation AI, and a collaborative environment and usage training were provided based on the Letcher platform for the entire organization to use.

The time and costs required during the planning and design phase have been drastically reduced, and collaboration between team members has also improved. As repetitive tasks have been reduced and real-time feedback has become possible, work speed has increased, and products currently planned with AI have reached the stage just before actual production.

We wanted to diversify and efficiently produce in-house broadcast content, but we were struggling to create differentiated content due to the lack of an AI voice that reflected natural pronunciation and brand identity like an actual announcer.

We planned and developed a high-quality TTS model based on Letser's AI platform “Lamp,” and created a custom voice that matches the brand persona. Global scalability, including multi-language functionality, and a sustainable update environment were built together to seamlessly integrate with the in-house platform.

Since various contents can be quickly produced based on AI voice, broadcast operation efficiency has been greatly improved. Member satisfaction was increased through content that strengthened brand identity, and iterative production resources were reduced and a stable operating environment was secured at the same time.

Askeye is a health care service based on traditional Chinese medicine that classifies patients' constitution and provides health information through iris analysis. However, the task of analyzing iris images with high accuracy was very exhausting for medical staff, and the lack of AI specialists made it difficult to adopt technology.

After refining image data held by medical institutions to be suitable for AI learning, we developed a model that automatically extracts key health-related information from iris images using computer vision-based AI. The results are linked to the service to provide a real-time analysis function, and a function to automatically provide initial diagnosis results to reduce the fatigue of medical staff has also been implemented.

Health information can now be extracted from iris images within 1 second, greatly reducing repeated analysis time for medical staff. The expertise and reliability of services has been improved by automating constitution information, and it is now possible to provide health care services tailored to individual patients.

Before the introduction of AI, thorough training was considered due to a lack of practical understanding of how to collect, refine, and manage vast amounts of data held by organizations and how it can be used for AI.

It provides practice-oriented training including understanding generative AI technology, learning data collection and quality control systems, and vector DB and LLM-based project examples. It is designed so that the training can understand the overall flow of data planning and AI model development.

We have secured basic knowledge and practical expertise for planning AI projects and proposed data management systems within the organization, and have laid the foundation for planning internal capabilities.

In a situation where new executives had to lead the AI business, it was difficult to plan a strategy in a situation where they lacked background knowledge about AI, and leadership-oriented training was needed to establish a clear direction.

We have organized a practice-oriented curriculum based on intuitive tools such as ChatGPT, Perplexity, and Intuitive AI, and designed so that executives can perform tasks using AI on their own, from planning prompts to establishing strategies and automating reports.

Executives were able to strengthen strategic thinking by directly utilizing AI in practice, and clarified the understanding and execution direction of AX projects within the organization through practice-based training.

In the process of deriving challenges for AI adoption, it was difficult to establish an actionable strategy because it was difficult to collaborate with various stakeholders within the organization and determine the feasibility of technology.

A design thinking-based workshop was conducted with Able Campus to structure the problems of each department and derive issues considering the applicability of AI. Expert engineers participated to review the technology and provide feedback on feasibility.

By redefining the overall business process, each group derived AI challenges that considered both feasibility and business value, and the trainees' understanding of collaboration and the perspective of applying AI in practice increased.

We wanted to introduce generative AI technology into QA work, but there was a lack of practice-oriented training to understand the super-large image generation model and internalize how to apply it to actual work.

Together with the Able campus, we designed a customized AI education curriculum tailored to the characteristics of the QA department, and also provided a Colab-based practice environment that non-experts can follow, and even practice understanding and performance evaluation of the LVM model.

By practicing the entire process from the principles to use of image generation AI, I developed a sense of technology that can be applied to actual QA work. The overall level of satisfaction among trainees was high, and their practical ability to select and evaluate AI models was strengthened.