Deccan AI Raises $25M, Hires Top Indian Talent
▼ Summary
– Deccan AI, a startup specializing in post-training AI data and evaluation services, has raised $25 million in a Series A funding round led by A91 Partners.
– The company provides services like improving AI coding capabilities and training systems to use APIs, primarily serving frontier AI labs and some enterprises.
– It employs about 125 people and utilizes a network of over 1 million contributors, with a major operations hub in Hyderabad, India, to manage quality.
– The market for these services is growing as AI labs outsource complex post-training work, where error tolerance is near zero due to impacts on model performance.
– Deccan has achieved a double-digit million-dollar revenue run rate, with about 80% of its revenue coming from its top five customers.
The demand for specialized services to refine and evaluate advanced AI models is fueling significant investment in a new generation of startups. Deccan AI, a company providing critical post-training data and evaluation work, has secured $25 million in a Series A funding round. This substantial investment, led by A91 Partners with participation from Susquehanna International Group and Prosus Ventures, underscores the growing need for external expertise as AI labs push their systems toward real-world reliability. Much of this intricate work is performed by a skilled, India-based workforce.
While leading AI developers like OpenAI and Anthropic build their core models internally, the subsequent phase of model refinement is increasingly outsourced. This work encompasses data generation, rigorous evaluation, and reinforcement learning. Founded in late 2024, Deccan operates in this space, offering services that help models improve at coding, function as autonomous agents, and interact with external tools like APIs. The company supports frontier labs with expert feedback, evaluation systems, and reinforcement learning environments, while also serving enterprise clients through its own products, such as the Helix evaluation suite.
Deccan’s business is growing alongside the evolution of the models themselves. As AI progresses beyond pure text into multimodal world models that understand physical environments for robotics and vision, the required training tasks become more complex. The startup currently serves about ten customers, including Google DeepMind and Snowflake, and manages several dozen active projects concurrently. Headquartered in the San Francisco Bay Area with a major operations center in Hyderabad, Deccan employs 125 full-time staff and leverages a network of over one million contributors, including students, domain experts, and PhD holders. In a typical month, between 5,000 and 10,000 of these contributors are active on the platform.
The broader market for AI training services is crowded, with established players like Scale AI and Surge AI competing with startups such as Turing and Mercor. What sets the post-training phase apart is an extremely low tolerance for error. Mistakes at this stage can directly degrade a model’s performance in live applications, making quality control paramount. According to founder Rukesh Reddy, achieving high-quality, domain-specific data at scale is a major challenge, compounded by intense time pressure. AI labs often require large volumes of accurate data within days, creating a difficult balance between speed and precision.
This sector has faced scrutiny over labor practices, given its historical reliance on gig workers for data tasks. Reddy states that earnings on Deccan’s platform vary widely, from approximately $10 to $700 per hour, with top contributors earning up to $7,000 monthly. About ten percent of the contributor base holds advanced degrees, a proportion that rises among actively engaged experts based on project demands.
India has become a central hub for this type of AI talent sourcing. Although Deccan’s primary customers are U. S.-based AI labs, the vast majority of its contributors are located in India. Competitors also tap into this talent pool but often spread their operations across numerous emerging markets. Reddy argues that concentrating its workforce in India allows for stricter quality management compared to a globally dispersed model. This strategy highlights India’s current role in the global AI ecosystem, primarily as a supplier of sophisticated training labor rather than a developer of the foundational models themselves.
Reddy describes Deccan as a “born GenAI” company, built from the ground up for the demands of generative AI, unlike older firms that adapted from computer vision labeling. This focus on higher-skill work from inception appears to be paying off. The company reports it grew tenfold over the past year and has reached a double-digit million-dollar annual revenue run rate. Approximately 80% of its revenue comes from its top five customers, reflecting the highly concentrated nature of the frontier AI market. While initially focused on India, Deccan has begun sourcing niche expertise from other regions, including the United States, for specialized fields like geospatial data and semiconductor design.
(Source: TechCrunch)
