High-quality mathematics training data for frontier AI models, created by IMO medalists and elite mathematicians.
Frontier AI labs struggle to obtain high-quality mathematics training data efficiently due to slow traditional data pipelines with multiple intermediaries, lengthy iteration cycles, and disconnects between data creators and researchers.
Frontier AI labs struggle to obtain high-quality mathematics training data efficiently due to slow traditional data pipelines with multiple intermediaries, lengthy iteration cycles, and disconnects between data creators and researchers.
Direct connection between frontier AI labs and elite mathematicians (IMO medalists, Putnam winners, PhDs) to design and create mathematics-focused reinforcement learning environments with rapid feedback loops and co-authorship partnerships.
Direct connection between frontier AI labs and elite mathematicians (IMO medalists, Putnam winners, PhDs) to design and create mathematics-focused reinforcement learning environments with rapid feedback loops and co-authorship partnerships.
Appears active as of March 2026 based on Y Combinator company listing and operational website.
Event Year: 2025
Appears active as of March 2026 based on Y Combinator company listing and operational website.
Event Year: 2025
hillclimb is a data company specializing in the creation of premium mathematics-focused reinforcement learning environments for artificial general intelligence development. Founded in 2025 by Jun Park and Ibrakhim Ustelbay, the company positions itself as a bridge between frontier AI research labs and the world's most talented mathematicians, enabling rapid advancement in AI model capabilities through superior training data quality.
Frontier AI labs require increasingly sophisticated training data to push model capabilities forward. Traditional data vendors operate through lengthy, inefficient pipelines involving multiple intermediaries, project managers, and outsourced teams. This approach creates significant delays—typically months between data requests and delivery—and prevents researchers from effectively iterating on data quality. By the time data arrives, model requirements have often evolved, making the dataset less valuable. Additionally, the disconnect between data creators and AI researchers means that the nuances of what makes training data effective for frontier models are often lost in translation.
hillclimb is a data company specializing in the creation of premium mathematics-focused reinforcement learning environments for artificial general intelligence development. Founded in 2025 by Jun Park and Ibrakhim Ustelbay, the company positions itself as a bridge between frontier AI research labs and the world's most talented mathematicians, enabling rapid advancement in AI model capabilities through superior training data quality.
Frontier AI labs require increasingly sophisticated training data to push model capabilities forward. Traditional data vendors operate through lengthy, inefficient pipelines involving multiple intermediaries, project managers, and outsourced teams. This approach creates significant delays—typically months between data requests and delivery—and prevents researchers from effectively iterating on data quality. By the time data arrives, model requirements have often evolved, making the dataset less valuable. Additionally, the disconnect between data creators and AI researchers means that the nuances of what makes training data effective for frontier models are often lost in translation.
B2B: Provides custom mathematics training data and reinforcement learning environments to frontier AI research labs on a project basis.
B2B: Provides custom mathematics training data and reinforcement learning environments to frontier AI research labs on a project basis.
Frontier AI research labs and large language model developers requiring high-quality mathematics training data for post-training and model improvement.
Frontier AI research labs and large language model developers requiring high-quality mathematics training data for post-training and model improvement.
Y Combinator company listing and operational website as of March 2026.
Hiring: unknown
Y Combinator company listing and operational website as of March 2026.
Hiring: unknown
hillclimb directly connects frontier AI research labs with a concentrated network of world-class mathematicians, including International Mathematical Olympiad medalists, Putnam competition winners, and research professors. This direct connection eliminates intermediaries and dramatically accelerates feedback loops, reducing iteration cycles from months to hours. The company's core offering is the design and creation of mathematics-focused reinforcement learning environments that serve as training data for large language models and other AI systems.
hillclimb directly connects frontier AI research labs with a concentrated network of world-class mathematicians, including International Mathematical Olympiad medalists, Putnam competition winners, and research professors. This direct connection eliminates intermediaries and dramatically accelerates feedback loops, reducing iteration cycles from months to hours. The company's core offering is the design and creation of mathematics-focused reinforcement learning environments that serve as training data for large language models and other AI systems.
The company operates on several key differentiators. First, it has assembled what it describes as the world's densest cluster of IMO medalists and elite math talent, providing unparalleled access to top-tier mathematical expertise. Second, it treats mathematicians as partners rather than gig workers, offering co-authorship opportunities on research papers when their data contributions become critical components of AI research. This partnership approach has reportedly achieved an 85 percent win rate in recruiting top talent against incumbent competitors.
The company operates on several key differentiators. First, it has assembled what it describes as the world's densest cluster of IMO medalists and elite math talent, providing unparalleled access to top-tier mathematical expertise. Second, it treats mathematicians as partners rather than gig workers, offering co-authorship opportunities on research papers when their data contributions become critical components of AI research. This partnership approach has reportedly achieved an 85 percent win rate in recruiting top talent against incumbent competitors.
Third, hillclimb maintains direct collaboration between researchers and mathematicians, allowing researchers to work within the community and see improvements in data quality within minutes rather than months. This tight iteration loop ensures that data quality remains consistently high because the people creating it understand exactly what frontier labs need.
Third, hillclimb maintains direct collaboration between researchers and mathematicians, allowing researchers to work within the community and see improvements in data quality within minutes rather than months. This tight iteration loop ensures that data quality remains consistently high because the people creating it understand exactly what frontier labs need.
Founders Jun Park and Ibrakhim Ustelbay bring complementary expertise to the venture. Park previously worked as a research engineer at DeepMind, giving him deep insight into what frontier AI researchers require from training data. Ustelbay founded Headstarter, an AI interviewer platform that scaled to over 100,000 users, demonstrating his ability to identify and vet top talent. Both founders transitioned from professional esports careers, teaching themselves to code and building successful technology ventures within a short timeframe.
Founders Jun Park and Ibrakhim Ustelbay bring complementary expertise to the venture. Park previously worked as a research engineer at DeepMind, giving him deep insight into what frontier AI researchers require from training data. Ustelbay founded Headstarter, an AI interviewer platform that scaled to over 100,000 users, demonstrating his ability to identify and vet top talent. Both founders transitioned from professional esports careers, teaching themselves to code and building successful technology ventures within a short timeframe.
As of early 2026, hillclimb operates with three employees based in San Francisco and is listed among the top reinforcement learning companies in the United States. The company has already begun partnerships with frontier AI labs, including work with Nous Research. The founding team's experience in both AI research and talent identification positions the company to scale its operations as demand for high-quality training data continues to grow.
As of early 2026, hillclimb operates with three employees based in San Francisco and is listed among the top reinforcement learning companies in the United States. The company has already begun partnerships with frontier AI labs, including work with Nous Research. The founding team's experience in both AI research and talent identification positions the company to scale its operations as demand for high-quality training data continues to grow.
hillclimb's long-term vision reflects a belief that the next era of AI progress depends on the quality of training data, which in turn reflects the intelligence and expertise of the people who create it. By assembling the world's brightest mathematical minds and connecting them directly with frontier labs, the company aims to accelerate the timeline toward artificial superintelligence while ensuring that the data driving this progress meets the highest standards of quality and rigor.
hillclimb's long-term vision reflects a belief that the next era of AI progress depends on the quality of training data, which in turn reflects the intelligence and expertise of the people who create it. By assembling the world's brightest mathematical minds and connecting them directly with frontier labs, the company aims to accelerate the timeline toward artificial superintelligence while ensuring that the data driving this progress meets the highest standards of quality and rigor.
