Aristotle University Of Thessaloniki (AUTH)
Μεθοδολογία μειωμένου κώδικα σχεδίασης και ανάπτυξης ασφαλών διαλογικών βοηθών για χρήση σε περιβάλλοντα περιορισμένων πόρων
Abstract
dc:descriptionIn recent years, there has been a rapid increase in the adoption and use of Conversational Assistants (CAs) and systems, especially after the emergence of Large Language Models (LLMs). These assistants are used in corporate environments to automate processes and search for corporate information, as well as by everyday users in smart homes and devices. Despite their impressive capabilities, their use is accompanied by significant challenges. The frequent occurrence of hallucinations, data security and ownership concerns, reduced results explainability and transparency, and increased hardware requirements are just some of the most significant challenges of these solutions. At the same time, the development of CAs remains a complex and time-consuming process, with high demands on technical knowledge, limiting their development to domain experts.This PhD thesis aims to develop a comprehensive methodology for the design and development of privacy-aware Conversational Assistants that can operate in low-resource environments. The proposed approach combines Model-driven Engineering (MDE) techniques, Domain-specific Languages (DSLs), and LLM-based conversational interfaces to automate the creation of CAs, which can be easily adapted to different use case scenarios, operate transparently, and be fully deployed in low-resource environments, such as edge devices. It focuses on the development and utilization of open hardware and software tools that ensure privacy, security, and adaptability. The PhD thesis is organized around four main objectives. Firstly, aims to define the processes and methodology for designing and developing CAs, thus creating a reusable and extensible CA development framework. Secondly, attempts to automate the CA development process by proposing several low-code tools, allowing even non-experts, such as citizen developers, to create conversational systems. Thirdly, focuses on optimizing the CA performance and particularly the Natural Language Understanding (NLU) component by proposing a methodology for selecting appropriate Language Models and algorithms. Finally, it explores the integration of LLM-based systems into the CA design and development process. This way, this PhD thesis contributes to the creation of a complete and practically applicable methodology for privacy-aware, explainable, and customizable Conversational Assistants, while offering tools that enable their development by a wider audience beyond domain experts.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Malamas, Nikolaos
- Μάλαμας, Νικόλαος
Subjects
dc:subject × 14- Διαλογικοί βοηθοί
- Μηχανική λογισμικού
- Τεχνητή νοημοσύνη
- Ανάπτυξη λογισμικού μειωμένου κώδικα
- Conversational assistants
- Software engineering
- Artificial intelligence
- Low-code software development
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Υπολογιστές, Υλικό (hardware) και Αρχιτεκτονική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Computer science, Hardware and Architecture
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60489