Vidraft, a deep tech startup specializing in artificial intelligence and scientific research, is making significant strides in developing advanced AI solutions. The company is focusing on creating a quantum operating system (OS) for quantum computers and a platform for the independent development of new drugs and materials. This dual approach positions Vidraft at the forefront of cutting-edge technological advancements, aiming to build robust AI research and development capabilities.
Vidraft’s Technological Foundations
Established in 2024, Vidraft has rapidly built a reputation for its innovative work. The company’s core offerings include the ‘Darwin Factory,’ a service designed to rapidly create specialized AI models, and its proprietary foundational model, ‘AETHER’. Vidraft is also developing the ‘Darwin’ series of models, which leverage model exchange and evolution techniques. These developments are supported by access to high-performance computing infrastructure, including NVIDIA’s Blackwhale GPUs, facilitated through participation in government-sponsored support programs.
Vidraft has achieved notable recognition for its technical prowess. It is the fourth startup in South Korea to surpass one million downloads of its self-developed large language model (LLM) on Hugging Face. Furthermore, the company secured first place in both the K-AI Leaders Board, operated by the Ministry of Science and ICT and the National Information Society Agency (NIA), and Google’s ‘The Fast Gemma Challenge’. These achievements underscore the company’s strong performance not only in model capabilities but also in practical AI implementation, including inference optimization and on-device serving using CPU-based infrastructure.
The Darwin Factory: Accelerating AI Model Development
The flagship solution, ‘Darwin Factory’, offers an AI Foundry service that combines two pre-verified open-source AI models to create customized, specialized AI models for client tasks. A key advantage of this service is its ability to generate a tailored model within a single day, eliminating the need for extensive GPU retraining or data collection.
The Darwin Factory service operates in three distinct stages:
- Diagnosis: This initial phase involves analyzing the hierarchical relevance of responses from open-source models to identify their respective strengths and areas of expertise.
- Exchange: Here, the strengths of two selected models are combined to create a single, integrated model.
- Verification: The newly formed model undergoes a proprietary benchmarking process to confirm its performance improvements over existing models. If the performance does not meet the required standards, the combination is discarded, and alternative pairings are explored.
Target Markets and Strategic Expansion
Vidraft is strategically targeting startups and mid-sized enterprises that lack in-house AI development capabilities. A significant focus is also placed on sectors where data exfiltration is strictly prohibited, such as defense, public administration, and finance. These industries often require AI solutions to be built within their own secure server or closed-network environments due to the sensitive nature of their data, making external cloud API usage challenging.
The company plans to expand its business into these closed-network markets starting in 2027, when external API access is expected to become increasingly restricted. Vidraft holds nine patents, including methods for neural network inference diagnosis, multi-diagnosis systems, and advancements in progressive anti-hacking and treatment for diseases like cancer. These patents form the basis for planned expansions into the defense, public, and financial sectors.
Future Outlook and Vision
Vidraft has successfully raised 650 million won in seed funding. The company’s phased strategy involves establishing a revenue base in the private sector over the next five years before expanding into the highly competitive closed-network markets. By the second half of 2026, Vidraft aims to secure numerous contracts with startups and mid-sized enterprises lacking AI development resources.
Starting in 2027, Vidraft will target the defense, public, and financial sectors, where data transfer restrictions necessitate on-premise or closed-network AI solutions. Key milestones for 2027 include converting initial engagements, such as Proof of Concepts (PoCs) with NH NongHyup Bank, civilian demand prediction RFPs with Hwaseong City and NIA, and proposals for the K-Defense industry growth project, into firm contracts.
Product enhancement is also a priority. Vidraft plans to commercialize ‘Darwin Factory v2’, which will automate the model creation process and establish standardized packaging for closed-network deployments, including installation, verification, and maintenance protocols. In 2028, the company intends to launch ‘AETHER’, a next-generation architecture designed for model integration, which will fundamentally replace current combination methods. Future expansion plans also include entering the Japanese and Middle Eastern markets.
Minsik Kim, CEO of Vidraft, articulated the company’s mission: “There are many companies that want to use AI but cannot due to lack of funds, personnel, or the inability to send data externally. We aim to create a world where these companies can have their own AI within their own servers.” He added, “We want to build a structure where businesses are not swayed by monthly usage fees or changes in external service providers’ policies. We aspire to become a company that produces our nation’s AI models, much like TSMC produces semiconductors.”
