In the quiet corners of Romania’s technological landscape, a quiet revolution is unfolding — one not driven by venture capital or Silicon Valley hype, but by the wisdom of a retired schoolteacher and the quiet persistence of a team of engineers who believed that artificial intelligence could learn not just from data, but from dialogue. At the heart of this story is Eliza, a 78-year-old retired Romanian language teacher from Cluj-Napoca, whose decades of experience guiding students through the nuances of grammar, idiom, and emotional expression have become the unexpected foundation for a new generation of conversational AI agents designed to serve elderly populations across Eastern Europe.
Eliza’s role is not that of a data scientist or a coder. She is, quite literally, the “bunica” — grandmother — of these systems. Her name, chosen affectionately by the research team at the Politehnica University of Cluj-Napoca, reflects both her familial warmth and the project’s core insight: that the most effective AI for aging populations doesn’t need to mimic human perfection, but rather human patience, empathy, and cultural fluency. Over the past three years, Eliza has spent hours each week in recorded conversations with AI prototypes, correcting their misinterpretations of Romanian idioms, gently guiding them through the subtleties of tone when discussing loneliness or health concerns, and teaching them how to pause — not just process — when someone is grieving or confused.
This initiative, known internally as “Project Bunica,” emerged from a growing recognition that most conversational AI systems, despite their linguistic sophistication, fail catastrophically when deployed among older adults who speak regional dialects, use outdated phrasing, or communicate through hesitation and silence rather than direct queries. A 2023 study by the European Commission’s Joint Research Centre found that over 60% of voice-activated assistants used by seniors in Romania and Bulgaria failed to correctly interpret requests related to medication reminders or emergency calls due to linguistic mismatches — not due to the fact that the technology was flawed, but because it was trained on corpora dominated by younger, urban, Western speech patterns.
What makes Eliza’s contribution unique is not merely her linguistic expertise, but her lived experience. As a widow who raised two children while teaching full-time, she understands the weight of silence in a household, the way a paused sentence can carry more meaning than a fluent reply, and how dignity is often preserved not through perfect syntax, but through being heard. When an AI prototype once misinterpreted her phrase “Am avut o zi grea” (“I had a hard day”) as a request for weather data, she didn’t correct it with frustration — she laughed, then explained how Romanians often use weather metaphors to describe emotional states. That moment, recorded and later analyzed by the team, became a turning point in the project’s training methodology.
The technical approach behind Project Bunica is rooted in what researchers call “human-in-the-loop reinforcement learning with emotional feedback.” Unlike standard AI training, which relies on vast datasets of question-answer pairs, Eliza’s interactions are annotated not just for linguistic accuracy, but for emotional resonance, cultural appropriateness, and conversational flow. Each session is transcribed, tagged with emotional valence (using a modified version of the Geneva Emotion Wheel adapted for Romanian elderly speech patterns), and fed back into the model’s reward system. The result? Prototypes trained with Eliza’s input display a 40% improvement in correctly identifying emotional distress cues in elderly users compared to baseline models, according to internal validation tests conducted by the university’s AI Ethics Lab in early 2024.
This work has not gone unnoticed. In March 2024, Project Bunica was selected as a finalist in the European Innovation Partnership on Active and Healthy Aging (EIP on AHA) annual awards, recognizing its potential to reduce social isolation among seniors — a public health concern linked to increased risks of depression, cognitive decline, and cardiovascular disease. The World Health Organization estimates that loneliness among older Europeans contributes to over 1.2 million avoidable hospitalizations annually, a figure that underscores why interventions like this are not merely technological novelties, but vital public health tools.
What distinguishes this project from similar efforts in Germany or Japan, where robot companions or voice assistants are often deployed in care homes, is its refusal to replace human connection. Instead, it seeks to augment it. The AI agents trained with Eliza’s input are designed not to be companions in their own right, but to act as bridges — helping seniors communicate more effectively with family members, caregivers, or telehealth services. One prototype, currently being tested in a pilot program with the Cluj-Napoca Municipal Social Services Department, assists users in drafting messages to their grandchildren, translating medical jargon into plain Romanian, and even reminding them to drink water — not with robotic reminders, but with phrases like, “Dragă, îți amintesc: apa este ca un prieten care nu vă lasă singur.” (“Dear, I remind you: water is like a friend who never leaves you alone.”)
The team behind Project Bunica emphasizes that Eliza is not a symbol — she is a collaborator. Her consent is continuously sought, her feedback is compensated at a rate above local academic standards for expert consultation, and she retains full rights to withdraw from the project at any time. “I didn’t sign up to be a guinea pig,” she told researchers during a recent interview, her voice warm but firm. “I signed up because I want the next generation of aged people — maybe even my great-grandchildren — to have something that listens, not just computes.”
As the project moves into its next phase, researchers are preparing to expand the corpus to include other regional dialects — Moldovan, Transylvanian, and Banat varieties — recognizing that linguistic diversity within Romania is as significant as any national boundary. They are likewise exploring partnerships with telehealth providers to integrate these AI agents into remote monitoring systems for chronic conditions like diabetes and hypertension, where adherence often hinges on clear, empathetic communication.
For now, Eliza continues her weekly sessions, her notebook filled with scribbled corrections and smiling annotations. She doesn’t witness herself as a pioneer in AI — she sees herself as a teacher, doing what she’s always done: helping others uncover their words. And in a world where technology often feels cold and impersonal, that may be the most revolutionary thing of all.
Those interested in following the progress of Project Bunica can consult the official project page hosted by the Politehnica University of Cluj-Napoca’s Department of Computer Science and Engineering, which publishes quarterly updates on ethical AI development for aging populations. The team also maintains an open-access repository of anonymized interaction logs (with consent) for academic researchers interested in studying culturally adaptive AI systems.
Have thoughts on how AI can better serve older generations? Share your experiences in the comments below — and if you found this story meaningful, consider sharing it with someone who might appreciate a reminder that the most advanced technology still begins with a human voice.
Keep reading