D3.2 Hybrid-AI and self-supervised models

This deliverable reports the algorithms, software components, and experimental results developed in tasks T3.1–T3.3, together with their integrated codebase implementation. The work focusses on improving context‑aware dialogue systems by exploiting and controlling contextual information across turns, with an emphasis on both low‑risk and high‑risk application scenarios. The proposed methods are designed to enhance interpretability and trustworthiness, by making model behavior more transparent and controllable, thereby contributing to the ELOQUENCE project’s overarching goal of building hybrid LLM‑based conversational agents that are effective, reliable, and aligned with user needs.