My journey into Computer Science began at home. My older sister first introduced me to coding, and I still vividly remember her explaining object-oriented programming to me while I was in high school. My parents have also always been enthusiastic about technology, so curiosity around computers and emerging technologies naturally became part of my upbringing.

That curiosity grew during my Bachelor’s in Computer Science and Engineering at BNMIT, where I went far beyond the curriculum through hackathons, workshops, and technical challenges. One of the most defining experiences was reaching the finale of the Smart India Hackathon as a lead team member, where we built a Virtual Tourist Guide using machine learning, user preferences, and contextual data to address a challenge presented by the Government of Goa. I also worked on projects such as a gesture-recognition system, a SafeSOS application, and a drone designed as a “third eye” for visually impaired users.

Hackathons became one of the most valuable parts of my learning journey because they taught me how to quickly explore new technologies and apply them to real-world problems. Through workshops and technical training programs, I was introduced more deeply to AI and machine learning, and several of my projects received best-project recognition during final demonstrations. That was when my interest in AI/ML became much more intentional.

After graduation, I began my career as a Software Engineer at Trellix, where I built automation tools, dashboards, and software frameworks. Wanting to deepen my expertise in AI/ML, I later moved to the United States to pursue a Master’s in Computer Science at North Carolina State University, with coursework in Machine Learning, Neural Networks, Data Science, Statistics, and Data Structures.

Today at FedEx, my focus has evolved from learning individual algorithms to building AI/ML solutions at enterprise scale. I work with technologies such as AI agents, retrieval-augmented generation, and intelligent automation to solve real customer and business problems. What has remained constant throughout the journey is my curiosity and the drive to keep learning, experimenting, and turning emerging technologies into practical solutions.

Working on AI/ML solutions at enterprise scale at FedEx has shaped my technical thinking the most. It changed the way I approach technology—from asking, “Can we build this?” to asking, “Can we make it scalable, reliable, efficient, and genuinely useful in production?”

I have learned that a successful AI solution is never just about the model. Data quality, system architecture, inference performance, infrastructure, observability, and integration with business processes all matter.

Whether I am optimizing AI inference or building AI agents for incident triage, I try to understand the problem end-to-end rather than focus on a single component. I start with the customer need, then look across the system to identify where technology can create the greatest impact.

That mindset has also shaped how I approach emerging technologies such as Agentic AI and RAG. I am excited by innovation, but I believe enterprise AI must go beyond impressive prototypes—it needs to be measurable, trustworthy, scalable, and useful in the real world.

When the solution is not immediately obvious, I first try to simplify the problem before trying to solve it. I break it into smaller components, identify what is known versus uncertain, and make sure I understand the actual customer or business problem behind the technical challenge.

From there, I usually explore multiple approaches, validate assumptions with data, and build small proofs of concept rather than committing too early to one solution. I also believe strongly in learning through experimentation—many complex problems become much clearer once you start testing ideas and observing how the system behaves.

Another important part of my approach is collaboration. I actively seek perspectives from people with different expertise, especially when the problem spans areas such as AI, infrastructure, data, and software engineering.

Most importantly, I stay comfortable with not having the answer immediately. Some of my best solutions have come from continuously questioning assumptions, learning quickly, and iterating until the simplest, most reliable solution becomes clear.

The problem I was trying to solve is that in AI data centers, installed capacity does not always mean usable capacity. A rack may be physically installed, but it may still depend on power, cooling, network connectivity, storage, monitoring, safety checks, or commissioning before it is truly ready to support AI workloads.

My research explored how evidence-grounded multimodal LLM agents could help determine that real state of readiness by bringing together information from multiple sources—such as rack and thermal imagery, telemetry, topology, work orders, commissioning records, specifications, and operating procedures. The goal was not simply to have an LLM generate an answer, but to make the system reason from verifiable evidence, identify conflicting information, explain why capacity may not be ready, and recommend the next appropriate action while abstaining when the evidence is insufficient.

The work was selected among the top five papers at the IEEE conference, which gave me the opportunity to present the research to a broader technical audience. That recognition was especially meaningful because it reinforced the importance of the problem and the growing need for trustworthy, operationally grounded AI systems.

For me, the broader value of this work is in showing how LLM agents can move beyond conversational use cases and support complex operational decision-making—while still maintaining traceability, human oversight, and safety.

As AI systems become more autonomous, I believe reliability has to be designed into the system from the beginning rather than added as an afterthought.

For me, the first requirement is grounding. An AI agent should make decisions based on trusted data, enterprise knowledge, and verifiable evidence rather than relying solely on what a model generates. The second is explainability—we should be able to understand what information influenced a decision, what actions the agent took, and why.

Observability is equally important. Just as we monitor traditional software systems, AI agents need strong logging, evaluation, confidence measures, and clear ways to detect when something is going wrong. I also strongly believe in human-in-the-loop design, particularly for high-impact decisions. Autonomy should not mean removing human judgment; it should mean knowing when the system can act independently and when it should ask for help or abstain.

Finally, usefulness matters just as much as intelligence. A sophisticated agent has limited value if it does not integrate naturally into the workflows of the people it is meant to support.

The future of Agentic AI, in my view, will depend not on how autonomous we can make systems, but on how responsibly we can combine autonomy with evidence, transparency, guardrails, and human judgment.

One of the biggest gaps is that research shows what a model can do, while production requires proving that it can do it consistently, efficiently, and safely at scale.

In real-world environments, model accuracy is only one part of the equation. Data quality, latency, infrastructure cost, observability, security, failure handling, and integration with existing workflows are equally important.

Evaluation is another major gap. Beyond benchmark scores, production systems must handle edge cases, hallucinations, drift, uncertainty, and situations where the model should abstain or involve a human.

Ultimately, deploying AI at scale requires strong engineering around the model—continuous evaluation, guardrails, monitoring, and clear operational ownership. The goal is not just to build something impressive in a demo, but something people can trust, maintain, and derive measurable value from over time.

I usually look at maturity through three lenses: technical reliability, business value, and operational readiness.

A new AI technology may perform well in a proof of concept, but before moving it into a real-world solution, I want to understand how consistently it performs outside controlled conditions. That means testing accuracy, latency, failure modes, edge cases, security, cost, and how well it handles uncertainty.

The second question is whether it is solving a meaningful problem. I try to avoid adopting technology simply because it is new. There should be a clear customer or business need, measurable value, and a reason why the AI approach is better than a simpler alternative.

Finally, I look at whether the surrounding ecosystem is ready. Can we monitor it? Can we explain its behavior? Are there appropriate guardrails and human-in-the-loop mechanisms? Can the system be maintained and scaled over time?

For me, a technology is ready to move beyond experimentation when it stops being just impressive and starts becoming dependable, measurable, and useful in the environment where people actually need it.

My biggest advice would be: do not wait until you feel completely ready, because in AI/ML, the field evolves too quickly for anyone to know everything.

Start with strong fundamentals, but learn by building. Some of my fastest growth came from hackathons, workshops, research projects, and hands-on experimentation where I had to understand a new concept and immediately apply it to a real problem.

It is also important not to compare your beginning with someone else’s expertise. AI/ML can feel intimidating because the field spans mathematics, software engineering, data, infrastructure, and constantly evolving models. You do not need to master all of it at once. Pick an area that genuinely interests you, go deep enough to build something meaningful, and gradually expand from there.

I would also encourage women to actively seek mentors, communities, conferences, and opportunities to present their work. Sharing what you build not only strengthens your confidence, but also opens doors to conversations and opportunities you may not have anticipated.

Most importantly, stay curious and be willing to take on problems that initially feel beyond your comfort zone. Technical depth develops over time. Curiosity, consistency, and the willingness to keep learning are what make the biggest difference.

Looking ahead, I am particularly interested in the evolution of AI from individual models into intelligent, adaptive systems. Areas such as multi-agent orchestration, multimodal reasoning, advanced retrieval architectures, and autonomous workflows are especially exciting to me because they move AI beyond simple assistance toward systems that can reason, collaborate, and act across complex environments.

I am also very interested in adaptive inference and AI systems optimization—how we can dynamically choose models, allocate compute, and improve reasoning at inference time while balancing performance, latency, and cost. As AI systems become more capable, efficiency will become just as important as intelligence.

Another area I want to explore is how agents can continuously learn from feedback, evaluate their own performance, and improve over time while still operating within strong safety, governance, and human-oversight boundaries.

Ultimately, I want my work to contribute to the next generation of enterprise AI systems: systems that are multimodal, agentic, efficient, and capable of solving complex problems with greater autonomy, while remaining explainable and trustworthy. My goal is to help bridge the gap between frontier AI research and dependable real-world systems that create measurable impact.