MISSION
HISTORY
CLIENT STORIES
2025
Confidential
Config.Inc
Problem
Config faces challenges in efficiently allocating, scheduling, and managing GPU resources, leading to underutilization, increased operational costs, and performance bottlenecks for computationally intensive workloads
Solution
We suggested our service DeepAutoSpace.
Results
We have entered into a large-scale supply contract with Config Intelligence Inc, an AI startup based in Delaware, USA, to provide the DeepAutoSpace service.
2024
ScaleServe
LG Electronics
LG Electronics is a global leader in consumer electronics, home appliances, and mobile communications. The Exaone model is part of their growing portfolio in AI-powered technologies and robotics. Exaone refers to a robotic system or AI assistant designed to work within smart home or business environments, potentially integrating with other LG smart products. These products enhance convenience, efficiency, comfort, and environmental sustainability for consumers.
Problem
The Exaone model faces significant challenges when it comes to running in an on-device environment, primarily due to the inherent hardware limitations
Solution
We suggested compressing the model size through LLM pruning and quantization, and applying scaleserve method.
Results
With our advanced solutions, our technology ultimately enables fast, long-context inference with limited memory, ensuring high performance and efficiency in resource-constrained environments.
ScaleServe
VMonster
On-going
VMonster is a company working on AI-powered conversational tools, specifically chatbots and natural language processing solutions. Their products focus on enterprise-grade AI for automating tasks, improving customer service, and enhancing communication via advanced conversational AI technology
Problem
VMonster AI provides scalable solutions through products such as chatbot solutions, virtual assistants, customer service automation, and AI-powered analytics. However, they were facing the problem of lacking training and serving infrastructure
Solution
They needed a space to manage the crucial infrastructure required to run their intelligent solutions. We, DeepAuto, solved this major problem by providing them with a cost-efficient workspace and model serving.
Expected Results
Using our products, and as this project is still ongoing, we aim to reduce total training and serving costs, which is a crucial step in both running the business and completing the user journey.
2023
GenAIOps for text-2-image
Cheil
Cheil Worldwide is not just a traditional advertising agency; it has embraced the power of AI and digital innovation to enhance its marketing and advertising solutions. As a global leader in creative marketing, Cheil has been incorporating AI-driven technologies into its strategies, leveraging them for better targeting, personalized experiences, and predictive analytics.
Problem
The process of creating images for advertisements is currently too slow, leading to delays in campaign execution and limiting the ability to quickly adapt to changing market trends and customer preferences
Customized Solution
We suggested implementing an image generation framework that seamlessly integrates and merges multiple concepts with high accuracy.
Results
Our proposed solution enabled the creation of highly detailed and contextually relevant visuals while maintaining coherence and visual integrity, ensuring impactful and effective advertising materials. In conclusion, it significantly reduced the time and costs associated with creating images for proof-of-concepts.
Model Compression
StradVision
StradVision specializes in computer vision and autonomous driving solutions. They focus on developing vision-based software for autonomous vehicles, providing AI-driven perception solutions that enable cars to "see" and understand their surroundings. The company's core product is a vision AI software platform that processes camera data and helps self-driving vehicles navigate safely in real-world environments.
Problem
It is extremely challenging to optimize self-driving models across a wide range of hardware platforms, as processing capabilities, memory limitations, and performance characteristics vary depending on the device type.
Customized Solution
As a solution, we suggested device- and architecture-aware model compression, which tailors the compression techniques specifically to the unique hardware and architectural constraints of the target platform, optimizing the model’s performance and memory usage.
Results
Our compression techniques successfully reduced the expensive model optimization costs for each device by thoroughly analyzing and identifying the unique constraints, allowing us to prepare highly tailored and detailed solutions for each device’s requirements.
PEOPLE
CEO
Co-Founder
Wonyong Jeong
CTO
Co-Founder
Jason Shim
CSO
Tekseon Shin
CBO
Don Chang
Advisor
