Deccan AI builds post-training, evaluation, and deployment tools for enterprise AI models. The AI infrastructure startup has now raised $25 million in a round led by A91 Partners, with Susquehanna and existing backer Prosus Ventures also participating. Lots of companies can access strong models now, but far fewer can safely train, test, and run them inside real business workflows without things breaking. Founded in 2023 by Rukesh Reddy, the company is betting that this messy middle layer — between a foundation model and a usable enterprise system — is where a lot of the value will sit.
What is Deccan AI and how does it work?
Deccan AI is trying to sell enterprises a full post-training stack, not a single AI feature. Its portfolio now includes STARK RL envs, Helix evals, and EnterpriseOS agents. In plain English, that means one layer for training agents in realistic conditions, one for generating and managing evaluation data, and one for deploying those systems into operating workflows.
The most concrete piece is STARK RL envs — the STARK RL gym. It simulates enterprise servers, tools, permissions, latency, rate limits, and irreversible actions so an AI agent can learn inside a controlled environment before touching live systems. The setup includes tasks, verifiers, golden trajectories, a sandbox container, plug-and-play LLM endpoints, and a Python SDK for training and evaluation. That’s a lot more useful than a toy benchmark. Enterprise failures usually come from workflow edge cases, not just bad prompt wording.
Helix evals sits closer to the data problem. It’s a data-generation tool for building, managing, and scaling high-quality training data, and that lines up with Deccan’s broader platform emphasis on expert-built datasets, model evaluation, domain-specific tuning for RAG, Text2SQL, coding, STEM, multimodal work, and agentic systems. The pitch isn’t “we’ll give you generic labels.” It’s “we’ll help you create the sort of evaluation and post-training data that enterprise models usually don’t have enough of.”
Then there’s EnterpriseOS agents. Deccan’s workflow for customers starts with understanding the business process and data sources. Then it customizes and trains a model on company data, and deploys and monitors it with a UI builder, sandbox testing, and real-time orchestration. Before that, a lot of this work lives in internal prompt hacks, manual QA, and scattered scripts. Afterward, the company is promising something closer to a managed production layer for enterprise AI. Ambitious? Yes. But the product logic is coherent.
Who founded Deccan AI and why now?
The founding story
Deccan AI was founded in 2023 by Rukesh Reddy. The company helps enterprises train, evaluate, and deploy AI across agentic workflows, coding, functional streams, and robotics — which explains why the new round is earmarked not just for post-training data and R&D, but also for enterprise-grade infrastructure and robotics-relevant data. It operates from the Bay Area, Hyderabad, and Bangalore. That fits the model: close to enterprise buyers in the US, deep talent delivery from India.
Why Rukesh Reddy fits this market
Reddy doesn’t come out of an academic AI lab. He comes from operating roles in finance and consulting — 15+ years across Citi, Monitor, and JPMorgan, with IIT Bombay and IIM Ahmedabad on the résumé. He also spent time at 360 ONE Wealth, where he led growth for the digital wealth business. That background matters because Deccan isn’t selling research demos. It’s selling reliability and process design. Enterprise trust, too.
Earlier operating experience
Before launching this company, Reddy held roles including SVP for strategy and business development in Citi’s global retail bank, US head of CX and digital transformation at Citi, and general manager for Citigold. He also founded Soul AI in 2023, another venture centered on RLHF and enterprise generative AI services. So while he isn’t a household-name model researcher, he does have a track record in complex operating environments where workflows, compliance, and customer experience are the whole game.
Early traction and signals
This isn’t pre-product vapor. Deccan AI already counts Google and Snowflake among its customers. The company has also built a talent pool of more than 500,000 specialists across 25+ domains for high-quality AI data and evaluation work — an important asset if your business depends on difficult post-training workflows rather than commodity annotation. It has also put enterprise certifications like SOC 2, ISO 27001, GDPR, and HIPAA front and center, which tells you exactly who it wants to sell to.
Funding details
The new round brings in $25 million, led by A91 Partners, with Susquehanna and Prosus Ventures participating. Deccan will use the money to scale post-training data, expand R&D, build enterprise-grade infrastructure, and deepen its datasets for enterprise use cases and robotics. That comes after Prosus had already backed the company in an earlier financing announced in May 2025.
Competition and positioning
This category is getting crowded fast. On the data and post-training side, enterprises can look at firms like Scale AI and Snorkel AI. On the evaluation side, buyers increasingly compare tools from Patronus AI, Arize, and Statsig, all of which focus on measuring model quality, production behavior, or guardrails in one form or another.
Deccan is trying to bundle the ugly parts together. Instead of only selling eval dashboards or only selling data services, it offers a chain from domain data creation to simulated RL training to live workflow deployment. Legacy alternatives are still messy — internal AI teams, outsourced contractor networks, systems integrators, and spreadsheet-heavy QA loops. Deccan’s bet is that enterprises would rather buy one stack that mirrors real operational failure modes than stitch together 4 vendors and hope the seams hold.
Why does Deccan AI's $25M round matter?
This isn’t growth capital for a simple SaaS seat-expansion story. The money is going into post-training data, R&D, and hardened infrastructure — the expensive stuff that determines whether an AI product survives contact with a real company. If Deccan executes well, it could move from being a useful vendor in model training and evals to something closer to a core enterprise AI plumbing layer.
For customers, that matters more than another flashy model demo. A lot of enterprise AI projects still fail in the handoff from benchmark to production. Deccan’s product set is built around that exact failure point. It trains agents on realistic workflows. It generates the right eval data, then deploys into live processes with monitoring and iteration. That’s a much less glamorous pitch than “we built a new model,” but it’s where many buyers are finally willing to spend.
For investors, the logic is pretty clear. Deccan already has known enterprise names on its customer list, a cross-border operating setup, and a product roadmap that maps neatly to where enterprise AI pain is heading. The hard part now isn’t whether there’s demand. It’s whether the company can scale quality without turning into just another labor-heavy services business wearing an infrastructure label.
Why are investors betting on AI post-training now?
The market tailwind is real. Gartner forecast worldwide generative AI spending at $644 billion in 2025, up 76.4% from 2024, and put software GenAI spending at $299 billion in 2025 with a path to $895 billion by 2028. That doesn’t mean every startup wins. It does mean the budget line is no longer theoretical.
Adoption is also getting broad enough that quality problems can’t be brushed aside as “pilot noise”. McKinsey’s 2025 global survey found 88% of respondents said their organizations were using AI in at least one business function, up from 78% a year earlier. But only about one-third said their companies had begun scaling AI programs, and just 23% reported scaling an agentic AI system somewhere in the business. That gap — lots of usage, much less dependable scale — is exactly where post-training, evals, and production workflow tooling become valuable.
There’s another shift underneath all this. Enterprises are getting less excited by raw model access and more obsessed with accuracy, governance, and workflow fit. Gartner even noted that many CIOs are growing dissatisfied with early proof-of-concept results and are leaning toward more predictable commercial solutions. So startups that can improve reliability after the model is chosen have a much clearer story than they did 18 months ago.
What should customers watch from Deccan AI next?
The thing to watch isn’t whether Deccan AI can add more product names to the site. It’s whether it can turn this three-part stack into a repeatable enterprise system with visible depth in a few verticals — especially robotics and other high-risk workflows where failure costs are real.
If that happens, this round will look smart.
If it doesn’t, Deccan AI risks getting squeezed between pure-play eval startups on one side and giant data infrastructure vendors on the other. That’s why the next 12 months matter so much. The company has money, customers, and a believable thesis. Now it has to prove the stack holds together at scale.
Read how Ultrahuman Secures ₹400 Crore in Series C Funding and why its smart ring-led health platform is taking on global wearable leaders.
FAQ
What is the latest Deccan AI funding round?
Deccan AI has raised $25 million in a round led by A91 Partners. Susquehanna and existing investor Prosus Ventures also joined, and the capital will be used for post-training data, R&D, enterprise infrastructure, and robotics-focused datasets.
How does Deccan AI work for enterprise customers?
Deccan AI combines training environments, evaluation tooling, and deployment software into one stack. A customer can simulate workflows in STARK RL envs, build higher-quality data and tests through Helix evals, and then push AI agents into operational systems through EnterpriseOS-style deployment tools.
Who founded Deccan AI?
Rukesh Reddy founded the company in 2023. His background spans Citi, Monitor, JPMorgan, and 360 ONE Wealth, and he studied at IIT Bombay and IIM Ahmedabad — which helps explain why Deccan’s pitch feels more enterprise-operations-heavy than research-lab-heavy.
Is Deccan AI an AI infrastructure startup or an AI services company?
It sits in an awkward but interesting middle ground. Deccan AI looks like an AI infrastructure startup because it sells productized tooling for post-training, evaluation, and deployment, but its human-expert data engine is also a big part of the value. That hybrid model could be a strength if customers want outcomes, not just software.
