Mercor Deccan AI competition raises $ 25M, experts from India

As the demand grows for training and refining AI models, Deccan AI – a startup providing post-educational information and analysis services – has raised $25 million in its first round of funding, with much of the work being done by Indian experts.

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

Although frontier AI laboratories including OpenAI and Anthropic build large models in-house, much of the graduate work – from data generation to evaluation and reinforcement learning – is increasing rapidly as companies push to create reliable systems for real-world applications. Deccan is seen as one of the new initiatives in demand.

Launched in October 2024, Deccan offers services ranging from supporting models to improve coding skills and agents to training machines to communicate with external tools such as application programming interfaces (APIs), which connect AI models to software applications.

The startup works with frontier labs on tasks such as providing expert solutions, evaluating and building inspiring learning environments, and supports businesses through products that include analytics, Helix, and an automation platform. The work is also progressing as species move beyond the literature on so-called “universal models” that better understand how the environment works, including robotics and visual systems.

Deccan’s clients include Google DeepMind and Snowflake, according to the company. It has grown to about 10 clients and runs several active applications at any given time, founder Rukesh Reddy (pictured above) said in an interview.

The startup, which is based in the San Francisco Bay Area with a large workforce in Hyderabad, employs about 125 people and relies on a team of over 1 million people, including students, area professionals, and PhDs. About 5,000 to 10,000 contributors work in a typical month, Reddy told TechCrunch.

About 10% of Deccan’s contributors have advanced degrees such as masters and PhDs, although that percentage is higher among contributors depending on the project’s needs, Reddy said.

The AI ​​education market has it growing rapidly along with the rise of major linguistic models, and companies such as Meta-owned Scale AI and his enemy The AI ​​is visibleand the basics Turing and Mercor Competing to provide documentation, evaluation, and training promotion.

“The condition remains an unsolved problem,” said Reddy, adding that tolerance for mistakes after training is “close to zero” because mistakes can directly affect the role of the model in production. This makes posterior studies more difficult than earlier stages, which require more precise, domain-specific data that are difficult to scale.

The work is also time-consuming, he said, with AI labs sometimes requiring high volumes of data within days, making it difficult to control speed and accuracy.

This part is done he was criticized for work and salarywith large pools of gig workers that are often used to create training information. Reddy said that the earnings on the Deccan platform range from $10 to $700 an hour, while top service providers earn $7,000 a month.

India emerges as a hub for AI training talent

While its customers are US-based AI labs, most of Deccan’s contributors are based in India. Competitors such as Turing and Mercor as well as source contracts from the country, but works across a many emerging markets.

Deccan chose to focus its workforce in India for better management, Reddy said. “Many of our competitors travel to 100-plus countries to find professionals,” he said. If you only work in one country, it’s easier to do well.

This strategy shows India’s position in the world’s AI value chain – as a supplier of talent and training data rather than a frontier model, which remains among a handful of US companies and a handful of Chinese players.

However, Reddy said Deccan has begun to source talent from a few other markets, including the US, for geospatial data and semiconductor design.

Reddy said Deccan was built as a born “GenAI” company, unlike other data-mining companies that started with computer services. This means that it has focused on high performance from the start.

Deccan has grown 10x in the last year and is now at the multi-million dollar level, said Reddy, refusing to share specifics. About 80% of its revenue comes from its top five customers, showing how frontier the AI ​​market is, he added.