Mallikarjuna Yadav Bachu.
Sixteen research studies, a newly granted U.S. patent and an AI baggage system already saving money on the ground: the engineer's work shows what happens when agentic AI meets real operations.
When a flight is cancelled at a major hub, thousands of decisions follow in minutes. Passengers need new itineraries. Bags must be rerouted. Gates, crews and ground vehicles have to be reassigned. For decades, those decisions have depended on people working phones and screens under enormous pressure. Mallikarjuna Yadav Bachu's research is built around a different idea: that AI agents can carry much of that load, as long as they can be trusted, measured and held to account.
Bachu, an AI and product engineering leader with 16 years of experience, has published 16 research studies since 2019. His most recent work, released this July, focuses on observability for autonomous agents: how to trace what an agent did, classify the ways it fails, and set reliability targets for multi-step agent workflows. In August, he was named co-inventor on a granted U.S. patent for real-time spoken translation between service agents and customers.
THE HEADLINE NUMBERS
On the ground. Bachu's work is not confined to papers. He led development of an AI-driven routing and dispatch system for baggage at one of the world's busiest airports, which handles roughly 80,000 bags a day. The system combines machine learning predictions of offload and transit times with bag cutoffs and triage rules, and re-optimizes driver routes every two minutes.
The results. Early results are striking: a 10 percent reduction in missed baggage, early gains of around 30 percent in handling efficiency, and more than $1 million in cost savings ahead of wider rollout. A patent application covering the approach was published in September 2025.
In customers' hands. He also leads a consumer AI assistant that helps travelers manage flight disruptions, rebooking, cancellations and bag tracking inside a mobile app, and earlier built an enterprise chat assistant that reached more than 95 percent accuracy under a rigorous LLM-as-judge evaluation.
WHAT HIS RESEARCH COVERS
Bachu's 16 studies trace a clear line from operations to autonomy.
Operations and devices. His earliest work applied probabilistic machine learning to baggage timing, giving planners confidence ranges instead of guesses. Studies that followed modernized the frontline toolkit: automated app delivery to large device fleets, RFID tracking on handhelds in noisy airport environments, and a risk-based path for moving enterprise mobile apps to new platforms.
Language and identity. From 2022, he turned to speech. His research compared translation approaches across more than 80 languages, explored how a handheld can tell who is speaking, and showed how personal details can be redacted in real time before speech is translated. A study on secure sign-on for shared devices addressed identity on the frontline.
AI security. As generative AI spread, Bachu examined its risks: the provenance of open-weight language models, due diligence for AI vendors, and the trust boundaries of the Model Context Protocol, including tool poisoning and server impersonation.
Agents. Since 2025, his work has focused on agents. He benchmarked leading agent frameworks for enterprise readiness and latency, compared emerging protocols for multi-agent interoperability, studied agentic AI for end-to-end flight disruption rebooking at consumer scale, explored language-model agents that reason over optimization solvers for real-time airport decisions, and, most recently, set out how to observe and measure agent reliability.
WHY IT MATTERS
Bachu's central contribution is closing the gap between AI research and AI in operation. Many studies propose methods. Fewer follow them into a working system that moves real bags or serves real customers. His do both, and the results feed back into the next study.
His work also shows a consistent concern for safety and accountability. The security studies, the privacy research and the new observability framework all ask the same question in different ways: how can an organization know what its AI is doing, and catch it when it goes wrong? As companies hand more decisions to autonomous agents, that question is becoming central to the whole industry.
The impact extends well beyond aviation. Hospitals, logistics firms, retailers and transport operators face the same mix of time-critical decisions, multilingual customers and pressure to adopt AI responsibly. The methods Bachu has published, from bag timing models to agent reliability targets, transfer readily to those settings.
WHAT'S NEXT
With agentic AI moving from pilot projects into core operations, the next challenge is reliability at scale: agents that can be trusted to act, and systems that show clearly when they should not. Bachu's latest research points squarely at that problem. On current form, his work will remain a reference point for organizations trying to make autonomous AI both useful and accountable.
