With the increasing need to train and refine AI models, Deccan AI—a startup that provides post-training data and evaluation services—has secured $25 million in its first significant funding round. Much of this specialized work is performed by a skilled, India-based workforce. The all-equity Series A round was led by A91 Partners, with additional investment from Susquehanna International Group and Prosus Ventures.

While leading AI labs like OpenAI and Anthropic develop their core models internally, a growing portion of post-training tasks—from data generation to evaluation and reinforcement learning—is being outsourced. Companies are turning to external partners to ensure their AI systems perform reliably in real-world applications. Deccan is positioning itself as one of the new startups meeting this demand.

Founded in October 2024, Deccan offers services that help AI models enhance their coding and agent capabilities, as well as train systems to interact with external tools like application programming interfaces (APIs), which connect AI models to software. The startup collaborates with frontier labs on generating expert feedback, conducting evaluations, and building reinforcement learning environments. It also serves enterprise clients through products such as its evaluation suite, Helix, and an operations automation platform.

EMBED_PLACEHOLDER_0

The scope of this work is evolving as AI models advance beyond text to incorporate “world models” that better interpret physical environments, including robotics and vision systems. Deccan’s current clients include Google DeepMind and Snowflake. According to founder Rukesh Reddy, the company has onboarded approximately 10 customers and typically manages a couple dozen active projects at any given time.

Headquartered in the San Francisco Bay Area with a major operations team in Hyderabad, Deccan employs about 125 people. It also draws on a network of over 1 million contributors, including students, domain experts, and PhD holders. Reddy noted that around 10% of Deccan’s contributor base holds advanced degrees such as master’s or PhDs, though this proportion is higher among actively engaged contributors depending on project needs.

The market for AI training services has grown quickly alongside the rise of large language models. Companies like Meta-owned Scale AI and its rival Surge AI, along with startups Turing and Mercor, are competing to supply data labeling, evaluation, and reinforcement learning services. “Quality remains an unsolved problem,” Reddy emphasized, explaining that the tolerance for errors during post-training is “close to zero” because mistakes can directly impact model performance in production. This makes post-training more complex than earlier stages, requiring highly accurate, domain-specific data that is challenging to scale.

He added that the work is often time-sensitive, with AI labs sometimes needing large volumes of high-quality data within days, which complicates balancing speed with accuracy. The sector has faced scrutiny over working conditions and pay, as many platforms rely on gig workers to generate training data. On Deccan’s platform, contributor earnings range from about $10 to $700 per hour, with top performers earning up to $7,000 per month.

**India emerges as a hub for AI training talent**

Although most of Deccan’s customers are U.S.-based AI labs, the majority of its contributors are located in India. Competitors like Turing and Mercor also source contractors from India but operate across a wider range of emerging markets. Reddy explained that Deccan concentrates much of its workforce in India to maintain tighter quality control. “Many of our competitors go to 100-plus countries to find the experts,” he said. “If you have operations in just one country, it becomes far easier to maintain quality.”

This strategy underscores India’s current role in the global AI ecosystem—as a provider of talent and training data rather than a developer of frontier models, which remain dominated by a small group of U.S. companies and a few players in China. However, Reddy mentioned that Deccan has started sourcing talent from other markets, including the United States, for niche expertise in areas like geospatial data and semiconductor design.

Reddy described Deccan as a “born GenAI” company, distinguishing it from traditional data labeling firms that started with computer vision tasks. This means Deccan has focused on higher-skill work from the beginning. Over the past year, the startup has grown tenfold and now operates at a double-digit million-dollar annual revenue run rate, though Reddy declined to provide specific figures. He added that about 80% of its revenue comes from its top five customers, reflecting the concentrated nature of the frontier AI market.