Deccan AI raises $25M

Deccan AI said on April 14 that it raised a $25 million Series A to build what it calls “super accurate” artificial intelligence systems for labs and enterprises. (deccan.ai) The round was led by A91 Partners, with Susquehanna International Group and Prosus Ventures participating, according to the company. Deccan AI is based in Mountain View, California. (deccan.ai) (prnewswire.com) Deccan AI sells post-training data, model evaluation, reinforcement learning environments, managed workforces and monitoring tools. Those are the layers companies use after a base model is built to make it behave more reliably in real tasks. (deccan.ai) (prnewswire.com) Founder and chief executive Rukesh Reddy said the company is focused on accuracy rather than raw model intelligence. The company said human review can lift many systems from about 70-80% accuracy to 90-95%, a range it argues is needed for production use. (deccan.ai) (moneycontrol.com) The company said it will use the money to expand enterprise products, deepen work on robotics and browser agents, and hire machine learning engineers to build reinforcement learning environments in the Bay Area. (deccan.ai) (prnewswire.com) Deccan AI said it already works with frontier labs and enterprise customers including Google DeepMind, Snowflake and a top-five frontier lab it did not name. Bar and Bench also reported those customer ties in coverage of the financing. (deccan.ai) (barandbench.com) TechCrunch reported on March 25 that Deccan AI runs much of its expert workforce from India, with about 5,000 to 10,000 contributors active in a typical month. That places it in the same fast-growing market for model training, evaluation and reinforcement learning services as Scale AI, Surge AI, Turing and Mercor. (techcrunch.com) CNBC-TV18 reported on April 27 that co-founder Purva Bhandari said the company is targeting 100x growth over the coming years as enterprises move from pilot projects to larger deployments. The pitch to investors is that companies now pay for fewer hallucinations and better monitoring, not just larger models. (cnbctv18.com) (deccan.ai) For Deccan AI, the funding is a bet that the next layer of the artificial intelligence stack is not the model itself, but the systems that make the model dependable enough to use. (deccan.ai) (prnewswire.com)