In conversation with CIO&Leader, Aditya Vardhan and Madhav Sharma, founders of Vexoo.ai say most AI systems struggle with ‘hallucination’ and don’t suit India’s small manufacturing enterprises. They emphasize that Vexoo’s solutions are built for the real world.
When Aditya Vardhan and Madhav Sharma talk about their company Vexoo.ai, they speak less like starry-eyed entrepreneurs and more like engineers with a chip on their shoulder. Both have lived through India’s startup trenches, Aditya via advanced AI and neuroscience research in Europe and Madhav through ventures in gaming and creative platforms via advanced AI and neuroscience research in Europe. Today, their shared mission is clear: build an Indian AI company that can compete with the world’s largest players, not by going bigger, but by going smarter.
Breaking the Scaling Law
Vexoo’s latest small language model benchmarks report makes bold claims. Its Trailblazer 1B model a mere billion parameters, scored 87.1 on GSM8K math reasoning, edging out GPT-4, and posted 71.8 on MMLU, a benchmark typically dominated by models seventy times its size. On the GPQA Diamond scientific reasoning test, the same model matched the performance of industry leaders at 88%.
“Everyone assumes you need 175 billion parameters and massive GPU clusters,” Vardhan says. “We’ve proven you don’t.”
If the results hold up to independent verification, MARLIN (Multi-head Adaptive Reasoning Later Injection) could mark a rare disruption: small, cost-efficient models outperforming trillion-dollar giants.
A Market Hungry for Accuracy
Vexoo’s pitch is simple: accuracy over hype. Sharma explains, “Hallucinations are the Achilles’ heel of current LLMs. In sectors like healthcare or law, a fabricated answer is not just wrong it’s dangerous.”
By their own measurements, Vexoo’s models reduce hallucinations by 35–40% compared to GPT-5, while running on infrastructure light enough to operate from a laptop.
Privacy is another selling point: unlike many competitors, Vexoo allows users to upload thousands of documents and query them without sending sensitive data to foreign APIs. This product line DocChat and Playground has obvious enterprise appeal. “Think of it as Perplexity for your private documents,” says Vardhan.
The Business Case
The numbers are small but growing: 100,000 users today, expanding at 30% month-on-month. Vexoo is in the process of raising $20–30 million, betting that a capital injection will accelerate adoption in key industries.
The interest is there. Dell’s Asia Pacific CTO reportedly told the founders their technology is “two years ahead.” In India, government partnerships are already bringing Vexoo into pilot projects that could reach millions of end-users. Global context makes the timing sharp. AI investment hit $300 billion in 2024, but 95% of enterprise deployments fail to scale. Vexoo’s leadership believes accuracy and efficiency not sheer size will define the next wave of adoption.
Competing in a Post-ChatGPT World
The founders are candid about what they see as a cooling hype cycle. “ChatGPT habituated people to convenience,” says Sharma. “But the honeymoon is over. Users now realize it isn’t accurate enough. That’s where we come in.”
Their roadmap includes expanding MARLIN adapters, which act like boosters to improve existing models. “Think of nitrous in a race car,” Vardhan says. “Plug our adapter into a BMW or a WagonR, and both go faster.”
India’s DeepSeek Moment?
The ambition runs deeper than the product. Both founders frame Vexoo as part of India’s strategic independence in AI. “This is bigger than nuclear technology,” Sharma insists. “It’s a perception war. If we keep relying on foreign APIs, we’re just consumers. We want India to be a producer.”
That vision might sound lofty, but it resonates. In an industry dominated by American and Chinese giants, an Indian company claiming GPT-4-level benchmarks with a 1B model is bound to attract attention.
Whether the claims withstand scrutiny remains to be seen, what’s certain is that Vexoo.ai has carved a compelling narrative: small, accurate, hallucination-free AI as a counterpoint to today’s sprawling, costly LLMs. They stress that Vexoo’s solutions are built for the real world, and as they aim to attract investments to scale and serve a wider audience, they believe that in a market desperate for trust and efficiency, Vexoo.ai claims to have what it takes to earn a seat at the global table.