Korea/spotlight
Korea Deep Learning started with public-sector OCR. Its founder was 22.
Jihyun Kim founded the Seoul company in 2019. It now sells document AI to more than 80 clients and has a model that topped an OCR benchmark.
Most documents that run a bank or a ministry are not tidy. They are scanned contracts, invoices with odd layouts, forms filled in by hand. Korea Deep Learning, a Seoul company founded in 2019, builds software that reads them.
Its chief executive, Jihyun Kim, was 22 when she started it, according to VentureSquare. A profile on AI Market Watch says she was then studying at Kyung Hee University; that detail appears in only one of the sources we could open, and the company has not confirmed it to us.
From OCR to vision-language models
The company began narrowly. VentureSquare reports that it first supplied a deep-learning optical character recognition system to public institutions, which gave it an initial customer base. It later moved to a vision-language model (VLM) structure, which reads the visual and linguistic context of a page together rather than treating text as a separate layer.
Its main product, DEEP Agent, is described as an integrated document AI system. It analyses text, tables, formulas, shapes and layout, restores table hierarchies and the links between clauses, and is designed to connect with ERP, RPA and CRM systems. A second VentureSquare report says it classifies contracts, invoices and administrative documents, then extracts and verifies the key fields.
The numbers
The company’s own model, KDL Frontier, ranked first worldwide in the English category of the OCRBench v2 benchmark in March, scoring 68.1, VentureSquare reports. The article says this was ahead of models from Google Gemini, OpenAI and Nvidia at the time. It is a single benchmark, and a narrow one, but a notable result for a company of this size.
On the commercial side, the same report counts more than 80 clients, mainly in the public, financial, industrial and logistics sectors, and cumulative revenue above 10 billion won. Customers report cutting document-processing time by up to 96 per cent, a figure the article attributes to them rather than to an audit.
AI Market Watch puts the Series A at about 12 billion won, roughly $9 million, led by Translink Investment, with SV Investment, Korea Development Bank and IBK Industrial Bank among the other investors. It dates the close to January 2026; VentureSquare says 2025. The sources disagree, so we leave the date open. AI Market Watch adds that the company reached its revenue figure by bootstrapping before the round.
Recognition
Kim was named to the 2026 Forbes Korea 30 Under 30 in the Deep and Enterprise Tech category. VentureSquare says she is the first woman in that category and was recognised for commercialising VLM document understanding. In May the outlet reported her selection to the Forbes Asia 30 Under 30 2026 list, as the only Korean in its consumer and enterprise technology category.
What to watch
Document AI is unglamorous work, and that is partly the point. The buyers are institutions with large backlogs of paper and strict accuracy requirements. A founder who arrived there at 22, through public-sector contracts, has built a business on revenue before a large round. VentureSquare says the company is now accelerating its global push. The test will be whether a benchmark lead in English translates into customers outside Korea.
