← projects

DIPLOMAT

2026-03-01 [in-progress]
DPOLLMsDialogue SystemsPreference AlignmentNLP

DIPLOMAT is a research project under review at ACL ARR March 2026, co-authored with Rishikant Chigrupaattii, Priyanshu Priya, and Dr. Asif Ekbal.

Core Idea

We propose Dialogue-Span-Aware DPO, a fine-grained preference-alignment method for workplace negotiation dialogues. Instead of treating a whole response as one preference object, the method works across multi-turn spans where persuasion, politeness, and negotiation strategy actually unfold.

Dataset

The work introduces PROWESS, a 2,400-dialogue dataset annotated with negotiation, persuasion, and politeness strategies using a multi-agent LLM framework.