Mercor rival Deccan AI has raised $25 million, according to expert sources from India

As demand for training and improving AI models grows, Deccan AI – a startup that provides post-training data and evaluation work – has raised $25 million in its first major funding round, with much of this work being carried out by a workforce of experts in India.

The all-equity Series A round was led by A91 Partners, with participation from Susquehanna International Group and Prosus Ventures.

While leading AI labs, including OpenAI and Anthropic, build basic models in-house, much of the post-training work — from data generation to assessment and reinforcement learning — is increasingly outsourced as companies seek to make systems reliable in real-world use. Deccan is emerging as one of a new crop of startups serving this demand.

Founded in October 2024, Deccan offers services ranging from helping models improve coding and agent capabilities to training systems to interact with external tools such as application programming interfaces (APIs), which connect AI models to software systems.

The startup works with leading labs on tasks like generating expert feedback, conducting assessments, and building augmented learning environments, while also serving organizations with products including its assessment suite, Helix, and a process automation platform. The work is also evolving as models move beyond text to so-called “universal models” that better understand physical environments, including robots and vision systems.

Deccan’s clients include Google DeepMind and Snowflake, according to the company. The company has about 10 clients and runs dozens of active projects at any given time, its founder Rokesh Reddy (pictured above) said in an interview.

The startup, which is headquartered in the San Francisco Bay Area with a large operations team in Hyderabad, employs about 125 people and relies on a network of more than 1 million contributors, including students, domain experts and Ph.D.s. Between 5,000 and 10,000 contributors are active in a typical month, Reddy told TechCrunch.

About 10% of Deccan’s contributor base holds advanced degrees such as master’s and doctorates, although the share is higher among active contributors depending on the project requirements, Reddy said.

The market for AI training services has expanded rapidly alongside the emergence of large language models, with companies such as Meta-owned Scale AI and its competitor Surge AI, as well as startups Turing and Mercor competing to provide data classification, evaluation, and augmentation services.

“Quality remains an unsolved issue,” Reddy said, adding that post-training error tolerance is “close to zero” because errors can directly impact the model’s performance in production. This makes the post-training phase more complex than previous phases, requiring highly accurate, domain-specific data that is difficult to measure.

He added that the work is also very time-sensitive, with AI labs sometimes requiring large amounts of high-quality data within days, making it difficult to balance speed and accuracy.

The sector has faced criticism over working conditions and wages, with large groups of freelancers often used to generate training data. Earnings on the Deccan platform range from about $10 to $700 per hour, with top contributors earning up to $7,000 per month, Reddy said.

India is emerging as a hub for AI training talent

Although most of its clients are US-based AI labs, most of Deccan’s shareholders are based in India. Rivals, such as Touring and Mercur, also source contractors from the country, but they operate across a wide range of emerging markets.

Reddy said Deccan has chosen to focus a significant portion of its workforce in India to better manage quality. “Many of our competitors go to more than 100 countries to find experts,” he said. “If you only have operations in one country, it becomes much easier to maintain quality.”

This approach highlights India’s current position in the global AI value chain — as a supplier of talent data and training rather than frontier model development, which remains concentrated among a handful of US companies and a few players in China.

However, Reddy said Deccan has begun sourcing talent from a few other markets, including the US, for specialized expertise in geospatial data and semiconductor design.

Reddy said Deccan was founded as a “GenAI” company, unlike traditional data classification companies that started with computer vision tasks. This means that it focused on higher-skilled work from the beginning.

Deccan has grown 10-fold over the past year and is now at a double-digit revenue rate of $1 million, Reddy said, declining to divulge specific details. He added that about 80% of its revenue comes from its five largest customers, reflecting the concentrated nature of the frontier AI market.