Rukesh Reddy

Founder Deccan AI

Last Reviewed: September 23, 2026

Executive Summary

Rukesh Reddy is the founder of Deccan AI, a Mountain View, California company that sells the unglamorous input frontier AI models now depend on: carefully checked human data for post-training, reinforcement-learning environments, and evaluation suites for AI agents. He started the company in October 2024 after more than fifteen years in finance and consulting at Citi, Monitor and JP Morgan, and within eighteen months had customers including Google DeepMind and Snowflake and a $25 million Series A led by A91 Partners. The business is built on a US commercial front end and a large, largely India-based expert network, and its pitch rests on accuracy rather than volume.

Career Highlights

• Founded Deccan AI in October 2024; the company reports roughly 125 employees and offices in the Bay Area, Hyderabad and Bangalore.
• Raised investment from Prosus Ventures, announced in May 2025.
• Closed a $25 million all-equity Series A in March 2026, led by A91 Partners with Susquehanna International Group and Prosus Ventures participating.
• Grew the business roughly tenfold in the year before the Series A, reaching what TechCrunch described as a double-digit million-dollar revenue run rate.
• Spent more than 15 years across Citi, Monitor and JP Morgan before founding the company.

Professional Journey

Reddy studied at the Indian Institute of Technology Bombay and the Indian Institute of Management Ahmedabad, and spent the next fifteen-plus years in financial services and strategy consulting, at Citi, Monitor and JP Morgan. Deccan AI’s own biography of him stops at those three names, and it gives no titles or dates for those years.

He founded Deccan AI in October 2024. Large-model development had shifted by then: pre-training on scraped text was well established, and what differentiated models increasingly came afterwards, in supervised fine-tuning, reinforcement learning from human feedback and, later, training agents in simulated environments. Every stage needs human work that is expensive to get right, such as expert-written answers, graded comparisons, coding tasks and multi-step agent traces. Deccan’s product line maps onto that sequence. It supplies SFT and RLHF data, a set of RL environments it calls STARK, an agent evaluation suite called Helix, and EnterpriseOS for companies deploying agents themselves.

The operating model uses both geographies. The company is headquartered in the Bay Area, while its contributor network, which it puts at more than a million people with between five and ten thousand active in a given month, draws heavily on Indian graduates and PhDs. TechCrunch reported that customers are concentrated: about ten in total, with the top five accounting for 80% of revenue. Reddy has argued that this work tolerates almost no error. He told TechCrunch that tolerance for mistakes in post-training “is close to zero,” because a bad label can degrade the model it trains.

Prosus Ventures invested in May 2025. In March 2026 the company announced a $25 million Series A led by A91 Partners, with Susquehanna International Group joining and Prosus participating again. Reddy said the money would go to scaling post-training data, research and enterprise infrastructure, and presented the round as a bet on the industry moving past chatbots toward agents that handle high-stakes business logic.

Education

Indian Institute of Technology Bombay

Indian Institute of Management Ahmedabad

Deccan AI’s own site names both institutions. It does not state a degree or field for either.