Dr Esma Nafiye Polat

Dr Esma Nafiye Polat is a Research Collaborator at the Cambridge Service Alliance. Her research at CSA focuses on trustworthy multi-agent service AI for customer-service and operational contexts. She is particularly interested in how agentic systems can coordinate customer interaction, recommendation, service recovery, routing, escalation, and human decision support across multi-session service journeys. Her current research direction examines how customer context and memory can be represented, updated, and evaluated so that AI-enabled service systems remain context-sensitive, auditable, and appropriately bounded by human oversight.

Esma completed her PhD in Computer Engineering at Özyeğin University. Her doctoral research introduced and analysed the Turkish Couple Dialogue dataset for conversational sentiment analysis, comparing contextual and non-contextual modelling approaches in Turkish multi-turn dyadic conversations. Her work examined transformer-based models, large language models, prompt-based and fine-tuned approaches, embedding-based representations, Turkish morphological features, and statistical validation of model performance. Her broader academic work includes peer-reviewed research on conversational sentiment analysis, speech-based attachment-style detection in married couples, and supervised learning methods for natural language processing.

Alongside her academic research, Esma brings around fifteen years of industry experience across machine learning, data science, natural language processing, banking analytics, telecommunications, aviation, recommendation systems, and conversational AI. Her recent work has focused on production-oriented generative AI and agentic AI systems for customer service and operational workflows, particularly in high-governance banking and enterprise service environments.

She has designed agentic AI architectures involving LLM orchestration, retrieval-augmented generation, structured outputs, tool/API integration, deterministic validation, auditability, and human-in-the-loop decision support. Her recent work includes multi-agent CRM and meeting-assistance workflows, trustworthy financial explanation systems, controlled operational AI workflows, and customer-support automation. Earlier, she worked on agent profiling, call-transcript analysis, agent-customer matching, recommendation and reranking systems, and NLP/LLM-derived behavioural signals for improving customer-agent interactions.