Lekha Priyadarshini Bhan has spent 13 years working at the intersection of data, machine learning and enterprise technology—but her work today is centred on one of AI's most important transitions: moving Generative and Agentic AI from impressive prototypes into systems enterprises can actually trust.
As an AI Architect and technology leader, her work spans Agentic AI, RAG, GraphRAG, multi-agent architectures and LLM engineering, with a particular focus on what happens after the demo: How do you evaluate an agent's decisions? How do you trace what it did? How do you control its actions? And how do you build AI systems that remain reliable when they become part of real enterprise workflows?
That production-first perspective has shaped a career that combines enterprise AI architecture, research, technical thought leadership and knowledge sharing. Lekha is a recipient of the Best Research Paper Award at IEEE CCEM, has delivered more than 1,000 hours of AI and technology training, and has contributed as a speaker and panelist at forums including Grace Hopper Celebration India (GHCI), Women in Data Science (WiDS) and the Indian School of Business (ISB).
Alongside her technical work, she is passionate about increasing the visibility and participation of women in AI and emerging technology—not simply as users of AI, but as engineers, researchers, speakers and technology leaders helping shape what comes next.
At the centre of all these strands is one principle: building AI that does not merely work, but can be understood, measured, controlled and trusted.
Lekha’s work extensively covers enterprise knowledge intelligence, Information technology (IT) service management, decision-support systems, and AI-powered automation. These areas require more than simply connecting an application to a Large Language Model (LLM). Her architecture work connects LLMs, knowledge graphs, vector search, tool calling, and workflow orchestration. Evaluation frameworks, observability and governance also form part of the systems she designs.
The objective is to make AI applications more reliable and traceable in real-world environments. As organisations increasingly experiment with Generative AI, questions around accuracy, monitoring, control and accountability become as important as the underlying model. The production-focused approach has become a defining part of Lekha’s work.
Lekha’s professional background spans data and machine learning before expanding into Generative AI and advanced AI architecture. Her journey also reflects the need for continuous learning and adapting to changing times and technologies. She holds a PG Diploma in Data Science from IIIT Bangalore, an MBA from ICFAI National College, and a Bachelor’s degree in Commerce.
The combination of business education and specialised data science training has supported her work at the intersection of technology and enterprise requirements. Today, her areas of focus include Agentic AI, RAG, GraphRAG, multi-agent architectures, and LLM engineering.
Lekha has extensive expertise in Retrieval-Augmented Generation (RAG), which enables AI applications to retrieve relevant information from external knowledge sources before generating responses. This is particularly valuable in enterprise environments, where AI systems need to work with organisation-specific information.
Her work also extends to advanced retrieval approaches such as GraphRAG and knowledge-graph-based systems, where relationships between information can be incorporated into the retrieval process. These architectures are particularly relevant for enterprise knowledge intelligence when information is distributed across interconnected sources.
But technology is only one part of the equation. Lekha’s work also looks at evaluation, observability and governance, which are critical when these systems move from demonstrations into production environments.
Lekha's work also reflects the next stage in the evolution of enterprise AI: Agentic AI.
While conventional AI applications may respond to individual prompts, agentic systems can be designed to work through multi-step tasks, use tools, coordinate actions, and interact with different components of an enterprise workflow.
Her recent work includes enterprise Agentic AI and intelligent ITSM automation, bringing agent-based approaches into practical business and technology operations. In IT service management, for example, AI systems can support processes involving incidents, classification, information retrieval, and decision-making.
Lekha's work specifically includes incident intelligence and root-cause analysis, as well as LLM-based taxonomy classification.
These applications illustrate how Agentic AI can move beyond conversational interfaces and become part of operational workflows. Rather than treating an LLM as an isolated component, the architecture can incorporate agents, tools, enterprise data, and workflow automation.
Her broader work in multi-agent systems and agent orchestration includes architectures where specialised agents perform distinct functions such as retrieval, reasoning, classification and decision support within a coordinated workflow.
A key focus of her work is building bounded, production-ready agentic systems, with evaluation, observability, human oversight and governance embedded into the architecture. This enables agent actions and decisions to remain measurable, traceable and controllable as AI systems move into real enterprise workflows.
Research has remained an important part of Lekha’s career alongside engineering.
She received a Best Research Paper Award at IEEE CCEM, adding formal recognition to her work in the technology and AI space.
Her research and writing cover areas including RAG, Agentic AI, AI evaluation, LLM engineering, and production architecture. Through these efforts, she has also contributed to the wider conversation around how organisations can build and assess AI systems.
Her interest in evaluation is particularly relevant as AI applications become more complex. A system that produces an impressive demonstration is not necessarily ready for deployment. Measuring performance, monitoring behavior, and understanding limitations become important parts of the engineering process.
Knowledge sharing is another major part of Lekha’s work. She has delivered more than 1,000 hours of AI and technology training, helping professionals understand emerging technologies and apply them to practical engineering problems.
The training work reflects a broader interest in making AI knowledge accessible beyond specialist research and engineering circles. For professionals entering a rapidly changing technology landscape, practical understanding can be as important as familiarity with new terminology. Lekha’s training focuses on helping learners bridge that gap.
Lekha has contributed as a speaker and panelist across technology and academic forums, including Grace Hopper Celebration India (GHCI), Women in Data Science (WiDS), and the Indian School of Business (ISB), sharing her perspectives on AI, emerging technologies and their practical application.
She is particularly passionate about increasing the participation and visibility of women in AI and emerging technology. Her interest goes beyond encouraging more women to enter the field; she wants to see more women building AI systems, conducting research, contributing technical perspectives, speaking on industry platforms and progressing into technology leadership roles.
Alongside enterprise engineering, Lekha actively contributes to technical thought leadership around Agentic AI, RAG, GraphRAG, AI evaluation, LLM engineering and production AI architecture. Her writing focuses particularly on the engineering challenges involved in building, evaluating and deploying AI systems for practical enterprise use.
With close to a decade and a half of experience across data, machine learning and enterprise technology, her current work reflects how quickly the AI field has evolved, and how the demands of the role of an AI architect are increasingly tied to questions of reliability, evaluation, governance and practical deployment.
For Lekha, the challenge is not simply getting AI to work. It is building systems that can be understood, measured, controlled, and trusted when they are put to work in the real world.