Participated · Series B · Apr 2024

Founded in 2020 and headquartered in South Korea, Upstage builds large language models (including its SOLAR LLM) and a document processing solution called Document PATH. Its product suite focuses on converting unstructured data into actionable insights through document parsing and information extraction to automate enterprise workflows. The company targets industries such as insurance, healthcare, and finance. Following the Series C, Upstage achieved a valuation above 1 trillion won and plans to expand GPU infrastructure and hire global AI talent. It also intends to scale operations internationally, particularly across the U.S. and Japan. The article did not disclose revenue, user counts, or other operating metrics.

Led · Series B · Jan 2024

Founded in 2020, Rebellions designs AI chips focused on inference workloads and outsources manufacturing. The company has released AI infrastructure products named RebelRack and RebelPOD, described as production-ready and scalable inference units. Rebellions has raised $850 million to date and is valued at approximately $2.34 billion after the latest $400 million pre-IPO round. It closed a $124 million Series B in 2024 and a $250 million Series C in November prior to this round. The startup is pursuing aggressive international expansion, having established entities in the U.S., Japan, Saudi Arabia, and Taiwan and aiming to build partner ecosystems with cloud providers, government agencies, telecom operators, and Neoclouds. Management has indicated the company is preparing for an IPO later this year while scaling commercial deployments of its inference infrastructure.

Participated · Series B · Oct 2023

Moreh builds MoAI, an AI software platform that optimizes and enables AI models and applications to run on GPUs and other accelerators (including NPUs) without code changes. Its flagship software is compatible with PyTorch and TensorFlow and positioned as an alternative to Nvidia’s tooling for larger-scale AI infrastructure. The company says MoAI increases GPU throughput and memory efficiency; Moreh reported that AMD’s MI250 with MoAI showed 116% higher GPU throughput than Nvidia’s A100 and that KT observed performance and memory advantages versus Nvidia’s DGX. Moreh has customers and partners including telco KT and chipmaker AMD, began generating revenue in 2021, and targets approximately $30 million in revenue by the end of 2023. The startup has 70 employees and plans to use the Series B proceeds for R&D, product expansion, and hiring; it recently completed training a 211-billion-parameter Korean LLM that it plans to open-source later this year.

Led · Equity · Sep 2023

Mathpresso operates QANDA, an AI-powered learning app that uses advanced optical character recognition to recognize text and mathematical equations and provide in-depth solutions and personalized content. Its product suite includes the QANDA App, 1:N online school (QANDA Study), and 1:1 online tutoring (QANDA Tutor). The company leverages proprietary AI technology and data to power personalization and tutoring features. Mathpresso plans to accelerate development of an education-specific large language model and incorporate that LLM into its AI tutor and existing services, leveraging KT’s infrastructure. The company raised an $8M investment from KT, bringing total funding to approximately $130M. QANDA is used by over 10 million users monthly and holds a top-seat position in more than 20 countries. Mathpresso, founded in 2015 by Ray Lee and Jake Lee, operates Qanda, a mobile learning platform for K–12 students. Qanda can decipher mathematical equations and text queries from a photo and deliver answers. The app reaches more than 14 million students each month across more than 50 countries, with 87% of users outside South Korea concentrated in Vietnam, Indonesia, Thailand, and Japan. Mathpresso is headquartered in Seoul and also maintains offices in Tokyo, Hanoi, and Jakarta. Since founding it has raised more than $122 million in total funding. The company says it will use the new capital to improve its offerings and execution. Mathpresso operates QANDA, an AI-based learning app for K-12 students that provides detailed math solutions and personalized learning content. Founded in 2015 by co-CEOs Ray Lee and Jake Lee, QANDA launched in 2016 and added AI-based OCR in 2017 to recognize text and mathematical formulas. The company claims over 45 million registered users, about 12 million monthly active users across 50 countries, and more than 3 billion education data points; roughly 85% of users are outside Korea. QANDA processes approximately 10 million photos uploaded daily and supports seven languages with offices in Tokyo, Hanoi, Jakarta and Bangkok. Earlier this year Mathpresso introduced a premium subscription offering unlimited micro-video lectures and added a community feature to enhance collaborative learning. QANDA is available on Android, iOS, and web. Mathpresso operates Qanda, a K-12 mobile learning app that delivers instant answers and customized learning resources using OCR to recognize text and mathematical formulas. The app is supported by over 2.4 billion solution data and a self-developed search engine, and serves more than 9.8 million monthly active users across 50+ countries. Qanda currently offers seven languages (Korean, English, Spanish, Japanese, Vietnamese, Indonesian, and Thai) and Mathpresso maintains regional offices in Tokyo, Hanoi, Jakarta, and Bangkok. The company was founded in 2015 by co-CEOs Ray Lee and Jake Lee and is headquartered in Seoul. Mathpresso plans to strengthen its AI-based recommendation algorithms and develop localized business models for its offices in Indonesia and Thailand. The company recently completed a financing round to support these initiatives. Mathpresso develops Qanda, an app that provides math and science help by letting students photograph problems and receive solutions from the app or human tutors using AI-based OCR. The app is used by roughly a third of students in South Korea and has users in more than 50 countries after launches in Japan, Vietnam, Indonesia and Singapore. The company emphasizes AI-driven personalized learning and is pursuing overseas expansion. Financially, Mathpresso has raised a total of $21.2 million to date. Its most recent financing was a $14.5 million Series B, following a $5.3 million Series A at the end of the prior year. Investors in the latest round include Legend Capital, InterVest, NP Investments and Mirae Asset Venture Investment.

Participated · Series B · Dec 2022

Sonde Health builds a vocal biomarker platform that uses advanced audio signal processing, speech science and AI/machine learning to detect subtle vocal changes tied to physiology. The company maintains a health‑labeled voice dataset of over 1.2 million samples from more than 85,000 subjects across four continents. Sonde offers enterprise products and licenses its platform for telehealth, pharma, remote patient monitoring, and consumer/medical devices, and it can embed technology into device chipsets for passive monitoring. Its Health Checks include respiratory fitness (detects conditions like asthma from six seconds of voice; users engage 2+ times per week) and mental fitness (detects signs of depression and anxiety from 30 seconds of voice; users engage 3+ times per week). Sonde is developing additional vocal biomarkers for MCI, COPD and general health using a research‑build‑validation methodology with clinical partners. The company intends to deepen its respiratory and mental health monitoring, expand into new health conditions and geographies, and accelerate commercial growth using the new funding. Sonde Health is developing a voice-based technology platform that uses acoustic analysis of seconds of voice to generate vocal biomarkers and persistent brain, muscle, and respiratory health measurements. Its proprietary technology is designed to run on consumer devices such as smartphones and smart speakers to enable disease screening and management. The company reports it has collected voice and health data from over 10,000 subjects as part of ongoing platform validation. Sonde intends to expand its technology across multiple health conditions and device types and to accelerate validation into neurological, respiratory, and cardiovascular diseases as well as other health and wellness conditions. The company plans to commercialize its offerings to provide broader access to timely health information and support new models of precision and preventative medicine. The announcement was datelined Boston and notes that Sonde was co-founded by PureTech Health.