—
The traveler at the counter speaks Portuguese. The agent speaks English. Between them is a rebooked itinerary, a missed connection and a growing line. For years, moments like this have been settled with hand gestures, a phone call to an interpreter line, or a helpful bystander. Mallikarjuna Yadav Bachu has spent much of the past year asking whether the handheld device already in the agent’s hand could do better.
Bachu is an innovation engineer who builds and tests new technology for frontline airport operations, and the author of six research studies published since 2019. His work moves across machine learning, enterprise mobility and, most recently, speech technology. The common thread is a simple idea: the people doing the hardest jobs in an airport deserve tools that are as smart as the problems they face.
TWO LANGUAGES, ONE CONVERSATION
His two newest studies, both published this summer, focus on real-time speech translation for customer service. The first compares two competing approaches. In a cascaded system, speech is converted to text, the text is translated, and the result is spoken aloud. In an end-to-end system, a single model goes directly from speech in one language to speech or text in another.
Bachu’s study measures both across more than 80 languages, weighing accuracy against delay. It pays particular attention to what researchers call the long tail: the many languages with far less training data than English, Spanish or Mandarin. Those are often the languages spoken by the travelers who most need help, and the ones where systems tend to fail quietly.
The languages with the least data are often spoken by the travelers who most need help.
The second study tackles a problem that sounds trivial until it isn’t. In a two-way conversation captured on one device, the system must know who is speaking before it can decide which direction to translate. Bachu’s research examines lightweight speaker attribution methods that can run on a handheld, keeping the exchange natural instead of forcing both people to press buttons and take turns.
Together, the studies reflect work he has been doing hands-on: building voice-to-text and text-to-speech features with automatic language identification and speaker recognition, so frontline agents can hold a conversation in a customer’s own language on the device they already carry.
MACHINE LEARNING ON THE RAMP
Bachu’s interest in operational machine learning predates the translation work. His first study, published in 2019, applied probabilistic models to baggage. Instead of assuming that every bag takes the same time to come off an aircraft and reach its connection, the research treated those times as ranges with measurable uncertainty.
That shift lets planners judge which connecting bags are truly at risk and act on them first. It is the kind of thinking that turns a ramp from a place of educated guesses into one guided by data, and it points to a future where routing and dispatch decisions for ground crews can be optimized continuously.
THE DEVICE IN EVERY HAND
Much of Bachu’s research comes from years spent building enterprise mobile apps for frontline workers. He has worked across native and cross-platform development with Kotlin for Android, Swift for iOS, Xamarin and Xamarin.Forms, .NET MAUI, Ionic and Angular, delivering more than ten business applications. He has built single sign-on with single logout across multiple Android apps, created a shared component library and UI standards for an app portfolio, and led scanner integrations for rugged devices from several major hardware makers.
That experience shaped three of his studies. One explored how continuous integration and delivery pipelines can be linked to mobile device management platforms, so updates reach large device fleets automatically and safely. He has applied the same approach in practice using Jenkins with enterprise device management tools.
Another examined RFID on handheld devices in the radio-noisy environment of an airport, where metal, vehicles and competing signals make reliable tag reads difficult. A third offered a risk-weighted framework for moving large portfolios of Xamarin apps to .NET MAUI, Microsoft’s newly released successor platform, one application at a time rather than in a single risky rewrite.
The people doing the hardest jobs in an airport deserve tools as smart as the problems they faceÂ
WHY IT MATTERS
Taken together, Bachu’s six studies describe a frontline transformed. The handheld device stops being a simple scanner and becomes a translator, a tracker and a decision aid. Updates arrive without disruption. Data about bags and assets flows in real time. And when a traveler and an agent do not share a language, the conversation can still happen.
The impact reaches beyond aviation. Hospitals, retailers, transit systems and logistics firms face the same problems: multilingual customers, large fleets of shared devices, and operations where minutes matter. Bachu’s research offers tested approaches to each.
There is also a human dimension that is easy to miss in technical work. Being understood is a matter of dignity. A traveler who can explain a problem in their own language is calmer, better served and more likely to trust the people helping them. Frontline staff, for their part, gain confidence and save time.
Speech translation, in particular, is moving quickly from research labs into everyday tools. Bachu’s focus on accuracy, delay and the long tail of languages addresses the questions that decide whether those tools work for everyone or only for the majority. Six studies in, his work makes a strong case that the most meaningful advances in technology often happen not in headlines, but at a counter, on a ramp, in the hand of someone trying to help.
THE ROAD AHEAD
Hard questions remain, and the field is still working through them. Real-time translation must be fast enough that a conversation still feels like a conversation. It must handle accents, background noise and the clipped, practical language of a busy airport. And it must treat what travelers say with care, since a short exchange at a counter can include names, booking details and other personal information.
Shared devices raise their own challenges. A single handheld may pass between several workers in one shift, each of whom needs secure access to their own tools. Getting identity, security and convenience right on those devices is a problem the industry has only begun to solve.
Practical, frontline problems like these are where Bachu’s work has always started. With six studies already completed and a body of work that spans machine learning, mobile engineering and speech technology, he has positioned himself at the meeting point of three fields that are reshaping frontline work. For travelers who have ever struggled to be understood at an airport counter, that is welcome news.
—
