University of the Aegean
Συστηματική υποβοήθηση νεοφυών επιχειρήσεων με χρήση τεχνικών τεχνητής νοημοσύνης: θεωρητική τεκμηρίωση, εργαλεία και εφαρμογές
Abstract
dc:descriptionThis dissertation examines the use of Artificial Intelligence to support and assess early-stage start-ups through systematic methodologies. Early-stage ventures operate under high uncertainty, limited financial and human resources, and unequal access to specialised mentoring and assessment capabilities. Traditional support mechanisms, such as consultancy services, incubators and accelerator programmes, provide valuable knowledge and guidance; however, they are constrained by cost, availability and scalability. At the same time, the increasing use of Generative Artificial Intelligence (GenAI) and Large Language Models (LLMs) has created new potential for providing entrepreneurial support on demand. Nevertheless, GenAI does not automatically address the underlying problem. When used as mentors or autonomous evaluators, general-purpose models remain prone to hallucinations, rely on static training data, demonstrate sycophancy towards user assumptions, exhibit unstable scoring and often fail to adapt to the development stage and business context of the start-up being assessed. The objective of this dissertation is to design, develop and empirically evaluate an integrated computational artefact, Evaluator.GR, which provides systematic support to a start-up from initial idea formulation and validation to funding preparation and investor readiness. Artificial Intelligence is viewed not as a wholesale replacement for human mentors, evaluators or investors, but as a supportive infrastructure that can automate or support selected tasks while augmenting human judgement, with emphasis on evidence-based decision-making, explainability and responsible use. Design Science Research (DSR) was selected as the primary methodological framework. The study was structured around three interrelated cycles: Relevance, Design and Rigor. During the Relevance Cycle, a Systematic Rapid Evidence Assessment was conducted to examine how Generative AI is applied in start-up mentoring and evaluation and to identify recurring limitations. The research gaps identified at this stage were translated into functional and architectural requirements for Evaluator.GR during the Design Cycle. During the Rigor Cycle, the artefact was evaluated through a comparative study against Gemini 3 and ChatGPT 5.2 and through a pilot implementation involving Pre-seed and Seed-stage start-ups. The developmental process resulted in Evaluator.GR, a modular platform supporting start-up evaluation, strategic analysis and the production of entrepreneurial deliverables. Its architecture combines a system-level Mixture of Experts, a Triple Reviewer configuration and a Meta-Judge mechanism. It also uses stage-adaptive evaluation rubrics, Retrieval-Augmented Generation and web-search augmentation. Human oversight is incorporated through a triple-layer Human-in-the-Loop framework. These features were developed to reduce reliance on a single general-purpose language model, improve the grounding and traceability of generated outputs and maintain human accountability for critical decisions. The artefact was also developed according to compliance-by-design principles and informed by relevant requirements and risk-management practices derived from the EU AI Act and the NIST AI Risk Management Framework. The study drew on three complementary sources of evidence. The Systematic Rapid Evidence Assessment included 34 final records published between 2022 and 2025. The comparative study involved 139 evaluators, with 52 participants in the Greek-language group and 87 in the English-language group, who assessed Evaluator.GR, Gemini 3 and ChatGPT 5.2 across 14 dimensions. In addition, a 30-day pilot study was conducted with ten Pre-seed and Seed-stage start-ups. Evaluator.GR received an overall mean rating of 4.51 out of 5, compared with 3.64 for Gemini 3 and 2.37 for ChatGPT 5.2. Within the specific sample and evaluation framework, these findings suggest that Evaluator.GR achieved higher perceived quality, evidence grounding, business utility and reliability than the two general-purpose models. Qualitative findings from the pilot further indicated that the system could identify material business or technical weaknesses across the examined use cases and provide context-sensitive recommendations. They also identified cases in which human intervention, reconsideration or recalibration of the generated results remained necessary.The dissertation makes theoretical, methodological, technical and practical contributions. It connects research on GenAI-assisted start-up mentoring, LLM-as-a-Judge systems and Responsible AI through the Mentor–Evaluator Stack; integrates evidence synthesis, artefact development and empirical evaluation within Design Science Research; and presents a multi-agent, stage-adaptive and knowledge-augmented architecture for supporting human judgement in start-up evaluation. The findings, however, concern perceived quality, usefulness and reliability within the examined setting. They do not establish a causal relationship between the use of Evaluator.GR and subsequent funding, survival or long-term growth. Additional limitations include the short duration and limited scale of the pilot, reliance on specific model releases, rubric versions and available datasets, and the additional operational and cognitive burden introduced by structured human oversight.
Degree
thesis:*- Grantor dc:publisher
- University of the Aegean
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Davalas, Athanasios
- Δαβαλάς, Αθανάσιος
Subjects
dc:subject × 26- Παραγωγική τεχνητή νοημοσύνη
- Νεοφυείς επιχειρήσεις
- Έρευνα Επιστήμης του Σχεδιασμού
- Μεγάλο Γλωσσικό Μοντέλο ως Κριτής
- Πρακτορική τεχνητή νοημοσύνη
- Υπεύθυνη Τεχνητή Νοημοσύνη
- Generative Artificial Intelligence
- Startups
- Design science research
- LLM-as-a-Judge
- Agentic Artificial Intelligence
- Responsible Artificial Intelligence
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Αυτοματισμοί και Συστήματα ελέγχου
- Οικονομικά και Επιχειρήσεις
- Κοινωνικές Επιστήμες
- Επιστήμη λήψης αποφάσεων
- Διαχείριση Τεχνολογίας και Καινοτομίας
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Automation and control systems
- Economics and Business
- Social Sciences
- Decision Sciences
- Management of Technology and Innovation
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62773