As the need for instructing and perfecting AI models expands, Deccan AI — a nascent enterprise supplying post-training information and assessment efforts — has procured $25 million in its initial significant financing round. A substantial portion of this undertaking is carried out by a skilled team of experts based in India.
The entirely equity-based Series A funding was spearheaded by A91 Partners, with participation from Susquehanna International Group and Prosus Ventures.
While cutting-edge AI laboratories such as OpenAI and Anthropic develop core models internally, a considerable amount of the post-training activities — encompassing data generation, evaluation, and reinforcement learning — is progressively being outsourced. This trend occurs as companies strive to ensure systems are dependable for practical application. Deccan is emerging as one of several new ventures fulfilling this escalating requirement.
Established in October 2024, Deccan offers services ranging from assisting models in enhancing their coding and agent proficiencies to educating systems on how to interact with external utilities like application programming interfaces (APIs), which link AI models to software frameworks.
The fledgling company collaborates with pioneering labs on assignments such as generating expert commentary, conducting assessments, and constructing reinforcement learning environments. Concurrently, it serves businesses through offerings like its evaluation platform, Helix, and an operational automation suite. This work is also developing as models transition beyond text to what are termed “world models,” which possess a superior comprehension of physical surroundings, including robotics and visual systems.
Among Deccan’s clientele are Google DeepMind and Snowflake, according to the firm. Approximately 10 customers have been onboarded, and several dozen active initiatives are ongoing at any given moment, stated founder Rukesh Reddy (pictured above) during an interview.
The enterprise, whose headquarters are located in the San Francisco Bay Area with a substantial operational team in Hyderabad, employs around 125 individuals. It also leverages a network of over 1 million contributors, including scholars, subject matter authorities, and PhD holders. In a typical month, between 5,000 and 10,000 contributors are actively engaged, Reddy informed TechCrunch.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
Around 10% of Deccan’s contributor pool possesses advanced academic credentials such as master’s and PhDs, though this proportion is higher among active participants depending on the specific project demands, Reddy mentioned.
The sector for AI training offerings has undergone rapid expansion alongside the proliferation of large language models. Firms such as Meta-owned Scale AI and its competitor Surge AI, along with startups Turing and Mercor, are vying to deliver data labeling, assessment, and reinforcement learning provisions.
“Maintaining quality remains an unresolved issue,” Reddy observed, adding that the tolerance for inaccuracies in post-training procedures is “nearly zero,” as mistakes can directly impact model efficacy in production. This renders post-training more intricate than earlier stages, necessitating highly precise, domain-specific data that is more challenging to scale.
The endeavor is also exceptionally time-sensitive, he remarked, with AI laboratories occasionally requiring significant quantities of premium data within a matter of days, making it arduous to reconcile celerity with precision.
The industry has encountered critique concerning working circumstances and compensation, as vast pools of temporary workers are frequently utilized to generate training information. Reddy stated that earnings on Deccan’s platform vary from approximately $10 to $700 per hour, with top contributors potentially earning up to $7,000 monthly.
India Becomes a Core Center for AI Training Expertise
Even though its clientele primarily consists of U.S.-based AI laboratories, the majority of Deccan’s contributors are situated in India. Competitors such as Turing and Mercor also procure contractors from the nation but operate across a broader spectrum of developing economies.
Deccan opted to concentrate a large portion of its personnel in India to better oversee quality, Reddy explained. “Many of our rivals extend to over 100 countries to locate specialists,” he noted. “If you conduct operations in just a single nation, it becomes considerably simpler to uphold standards.”
This methodology underscores India’s current standing within the global AI value chain — serving as a provider of skilled individuals and training data rather than a creator of cutting-edge models, which largely remain concentrated among a select group of U.S. corporations and a few entities in China.
However, Reddy indicated that Deccan has commenced recruiting talent from a few other regions, including the U.S., for specialized knowledge in geospatial data and semiconductor architectural design.
Reddy stated that Deccan was conceptualized as an intrinsically “GenAI” firm, in contrast to conventional data labeling companies that initiated their operations with computer vision assignments. This implies it has concentrated on more highly skilled tasks from its inception.
Deccan expanded tenfold over the previous year and currently boasts a double-digit million-dollar revenue run rate, Reddy disclosed, choosing not to share precise figures. Roughly 80% of its income originates from its top five clients, reflecting the concentrated nature of the frontier AI marketplace, he further added.
{content}
Source: {feed_title}