Deccan AI raises $25 million to build high-accuracy AI systems for enterprises and frontier labs

Deccan AI Secures Major Funding for AI Training Services

Deccan AI raises $25 million in an all-equity Series A round led by A91 Partners, with participation from Susquehanna International Group and Prosus Ventures. The startup, founded in October 2024 by Rukesh Reddy, has quickly positioned itself as a critical player in the AI post-training market, serving frontier model labs and enterprises with specialized data generation, evaluation, and reinforcement learning services 1

1

. This marks A91 Partners' first investment in an artificial intelligence startup, signaling growing investor confidence in the AI infrastructure sector .

The company has onboarded about 10 customers and runs a couple of dozen active projects at any given time, with approximately 80% of revenue coming from its top five customers. Deccan AI grew 10x over the past year and now operates at a double-digit million-dollar revenue run rate, though specific figures were not disclosed 1

1

.

Building High-Accuracy AI Systems for Enterprise AI and Frontier Labs

Deccan AI provides services ranging from improving coding and agent capabilities to training systems that interact with external application programming interfaces (APIs). The startup works with frontier model labs on tasks such as generating expert feedback, running evaluations, and building reinforcement learning environments 1

1

. Its customers include Google DeepMind and Snowflake, reflecting the concentrated nature of the frontier AI market.

Rukesh Reddy emphasized that the industry has reached a critical juncture. "We've reached a tipping point where the industry is moving past the chatbot phase. Getting an agent to work in a demo is one thing; getting it to handle high-stakes business logic is another," he told ET 2

2

. While large language models have improved rapidly, their probabilistic nature makes them inconsistent in real-world deployments, especially in enterprise settings where the cost of error is high.

AI Model Evaluation and Enterprise Products Drive Growth

The company has developed two key enterprise offerings. Helix, a hybrid human-plus-automated evaluation suite, helps enterprises monitor whether AI models are performing reliably in production. EnterpriseOS focuses on automating back-office operations and middle-office workflows using AI agents deployed within client infrastructure 2

2

. This platform-based approach combines training data, reinforcement learning environments, and human expertise to improve models.

Six months ago, the business was entirely focused on frontier model training. It is now roughly 90% frontier and 10% enterprise, but the enterprise segment is expected to grow faster as AI deployment scales 2

2

. The company is primarily targeting Fortune 500 firms, with global capability centres serving as a key entry point.

India-Based Workforce Powers Quality Control

Headquartered in the San Francisco Bay Area with a large operations team in Hyderabad, Deccan AI employs about 125 people and relies on a network of more than 1 million contributors, including students, domain experts, and PhDs. Around 5,000 to 10,000 contributors are active in a typical month 1

1

. About 10% of Deccan's contributor base has advanced degrees such as master's and PhDs, though the share is higher among active contributors depending on project requirements.

Even as its customers are largely U.S.-based AI labs, most of Deccan's contributors are based in India. Reddy said the company chose to concentrate much of its workforce in India to better manage quality. "Many of our competitors go to 100-plus countries to find the experts. If you have operations in just one country, it becomes far easier to maintain quality," he explained 1

1

. Earnings on Deccan's platform range from about $10 to $700 per hour, with top contributors earning up to $7,000 a month.

Deccan AI is also setting up a Bengaluru office focused on enterprise business, complementing its existing presence in San Francisco and Hyderabad. The team in Bengaluru is expected to grow from a handful of employees to 20-30 over the year 2

2

.

Addressing Critical Quality Challenges in Post-Training

"Quality remains an unsolved problem," Reddy said, adding that tolerance for errors in post-training is "close to zero" as mistakes can directly affect model performance in production 1

1

. That makes post-training more complex than earlier stages, requiring highly accurate, domain-specific data that is harder to scale. The work is also highly time-sensitive, with AI labs sometimes requiring large volumes of high-quality data within days, making it difficult to balance speed with accuracy.

Kaushik Anand, partner at A91 Partners, said Deccan is building the essential infrastructure for the next decade of software. "As the world moves from experimentation to execution, the need for Deccan's evaluation and monitoring layer becomes non-negotiable. They are making it safe for consumers and enterprises to actually trust AI" 2

2

. The work is also evolving as models move beyond text into so-called "world models" that better understand physical environments, including robotics and vision systems.