.png){kind=logo}
“True progress in AI is not about what we can make machines do, but what we can make them understand.”
In the evolving world of artificial intelligence, progress is no longer measured only in parameters and speed. The conversation is moving toward reasoning, reliability, and the capacity to explain how conclusions are formed. Amid this shift, Vishvesh G. Bhat has shaped a path that connects scientific depth with ethical clarity.
From his early research at the University of California, San Diego to his ongoing work at CoreThink AI, Vishvesh’s journey traces a quiet conviction that intelligence, whether human or artificial, must be transparent to be trusted.
Vishvesh’s academic story began at during his undergraduate studies at UC San Diego. His research focused on Natural Language Inference and multilingual language modeling, particularly Arabic, where cultural nuance and linguistic diversity often exposed the limits of existing systems. He recalls those years as a study in how machines misinterpret context.
Models could predict patterns, but they often failed to grasp intent. That tension between precision and understanding became the central theme of his career. The work made him question whether AI systems could ever reflect human reasoning rather than simply mirror data.
The search for that balance between efficiency and empathy would later guide his approach to building explainable systems.
Moving beyond academia, Vishvesh began testing his ideas in applied environments that demanded both scale and practicality.
At Tangible AI, he helped develop Qary, an open-source conversational platform aimed at democratizing intelligent dialogue.
At Viva, his work supported real-time multilingual collaboration tools that helped people communicate across barriers of language and culture.
During his time at Primer Technologies, he built relation-extraction models that improved data efficiency by over forty percent. The achievement was technical, but its lesson was philosophical performance meant little if users could not understand the system’s reasoning.
At Twinly AI and Pebble Finance, he deepened this pursuit and experimented with multi-step logic chains and modular reasoning. He focused on improving factual accuracy in generative outputs, raising the reliability of automated financial reports.
Across these roles, a single idea persisted: that trust in technology begins with visibility.
Ultimately, this idea developed into a framework. In 2025, Vishvesh founded CoreThink AI as an experiment in structured transparency: a "space" that would allow neural reasoning and symbolic reasoning to work together instead of working apart. The method avoids black-box predictions, using instead traceable reasoning paths to show how each insight is produced. For Vishvesh, the aim was not speed, but accountability.
He often describes the company’s philosophy in simple terms: “An intelligent system should be able to explain itself.” This belief has quietly influenced how engineers and researchers think about logic, bias, and verification in modern AI.
Aside from CoreThink's development philosophy, Vishvesh has been outspoken about the broader cultural opportunities of explainable AI. Often, he is able to express that transparency is not merely a technical deliverable, but a mindset through which teams consider design, communication, and deployment of intelligent systems.
By promoting documentation, open validation and verification, and reproducible research, he has helped change the perception of AI as a closed field of practice and experiment - to one of collegial, transparent systems and accountability across teams. His public talks and conversations in knowledge-sharing circles have motivated a select few younger engineers to think of reasoning not as limitation, but as the basis of new technology.
In an interview with ODBMS, Vishvesh spoke about the evolution of what he calls Large Reasoning Models (LRMs) - systems that combine the pattern recognition of large language models with structured chains of logic.
He explained that while existing models can generate language convincingly, they often struggle with sustained reasoning. CoreThink’s approach integrates symbolic elements that allow each inference to be traced, tested, and validated. This creates a layer of accountability often missing from traditional AI systems.
In the conversation, he emphasized that progress in AI should not be measured only by what systems can predict, but by how clearly they can justify their predictions.
It was a reminder that transparency is as much a human responsibility as it is a technical one.
Vishvesh’s story offers a quiet counterpoint to the noise surrounding artificial intelligence.
It suggests that the next breakthroughs will not come from larger models, but from clearer thinking.
His pursuit of transparency reflects a broader shift in the field from scale to substance, from output to understanding. In making systems that can explain themselves, he is also reminding us what intelligence, at its best, truly means.
