Nums AI, a startup building foundation models for tabular data, closed the second tranche of its pre-seed round with USD 1.5 million (2 billion won) in investment.
Mirae Asset Venture Investment and Korea Investment Partners each invested USD 760,000 (1 billion won). Nums AI was also selected for TIPS (Tech Incubator Program for Startup), the Ministry of SMEs and Startups’ private-investment-led tech startup support program, and will receive USD 600,000 (800 million won) in research funding.
The company earlier raised its first investment of USD 3 million (4 billion won) in July from Stonebridge Ventures, SBVA, KT Investment and Bass Ventures. Nums AI was founded in February this year. It drew strong interest from the investment community by raising a large pre-seed round on the strength of its founding team’s technical capabilities alone.
The second closing brings cumulative funding to USD 4.6 million (6 billion won). Nums AI plans to use the proceeds to develop its tabular-data AI models and hire staff.
Nums AI was selected for the 2026 third-round TIPS R&D General Track, with Bass Ventures as the operating firm. The project’s research funding is USD 600,000 (800 million won). TIPS links private investors’ investment and incubation with government R&D support.
Nums AI’s Causilo is a foundation model for tabular data. It makes predictions from data organized in rows and columns, such as sales records, transaction histories and customer information. It can be applied to business decisions such as predicting customer churn or product sales volumes. The core aim is to connect the numerical data a company has accumulated to actual business decisions.
When unveiled in September 2026, Causilo ranked first in the overall, classification and regression evaluations of TabArena, a global benchmark for tabular-data AI. The company says the result shows its technical achievement, since it designed and pre-trained its own model for the characteristics of tabular data and prediction tasks, then validated its performance.
During Nvidia CEO Jensen Huang’s keynote at NVIDIA GTC Berlin 2026 in Berlin on October 21, 2026, Causilo was presented as a major global model alongside Nvidia’s Kumo-Tabular and TabPFN-3.5 from Germany’s Prior Labs. Causilo’s code and pre-trained model weights are publicly available, while commercial use requires a separate license.
“Speed matters more than anything in the increasingly competitive field of tabular-data AI,” said Jaemin Yoo, CEO of Nums AI. “We will accelerate model development with this follow-on investment.”
Nums AI was founded in February this year by researchers from Seoul National University and KAIST, led by Professor Jaemin Yoo of Seoul National University’s Department of Computer Science and Engineering. The company is working on projects with various Korean companies, including InBody. It is developing technology that connects tabular-data AI to on-the-ground prediction and decision-making.