04441ntm a22006137i 4500 000729253 CZ-PrVSE 20260315164443.0 m d cr n|||||||||| 260315s2025 xr fsbm 000 0 eng d NEZPRACOVANÝ IMPORT ABA006 cze ABA006 ABA006 rda Šmíd, Soňa ISIS:188199 dis AIbō : Design and Evaluation of an Ontology-Guided AI Sales Assistant for Partner Incentive Programs eng AIbō : Design and Evaluation of an Ontology-Guided AI Sales Assistant for Partner Incentive Programs Soňa Šmíd 2025 ?? stran : digital, PDF soubor Vedoucí práce: Jan Černý Závěrečná práce - Institut celoživotního vzdělávání (MBA)—Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky, 2026 Obsahuje bibliografii Textový (vysokoškolská kvalifikační práce) Rok obhajoby 2026 AIbō is a conversational AI assistant designed to support Adastra's sales and alliance teams in navigating complex partner incentive programs. It is powered by a Retrieval Augmented Generation (RAG) pipeline grounded in a vendor-agnostic ontology and connected to an OpenAI language model through a FastAPI backend and a Supabase PostgreSQL vector database. The objective of AIbō is to reduce manual administrative effort, improve the accuracy and speed of incentive discovery, and support revenue acceleration through more effective use of partner funding. This MBA project presents the design, implementation, and evaluation of an AIbō prototype built on a Lovable frontend, FastAPI backend, and Supabase database, with development support provided by Cursor IDE and an intended migration path to Agate, Adastra's enterprise GenAI platform. The prototype achieved measurable improvements in accuracy, citation reliability, and correct display of visibility tags (internal and external) in citations, validating ontology-guided retrieval as an effective approach for enterprise assistants. A key finding of the pilot was that ontology-backed responses reached 77.6% accuracy and 100% citation reliability, outperforming non-ontology queries by 34.3 percentage points. Web search functionality and partner portal integration were intentionally disabled during the pilot to focus on the robustness of RAG-only retrieval, ensuring that performance reflected the system's reasoning supported by low-temperature model settings aimed at factual consistency. In future iterations, AIbō will evolve toward agentic functionality, capable not only of retrieving information but also of proactively managing knowledge through change detection, escalation, and integration with partner portal APIs. The project contributes to applied research by demonstrating how ontology and RAG can be combined through vibe coding to create conversational systems that deliver tangible business value. Způsob přístupu: Internet Conversational AI Retrieval Augmented Generation (RAG) Partner Incentive Programs Hyperscalers Ontology Large Language Models (LLMs) Cursor Lovable Supabase Pilot Evaluation Sales Enablement Vibe Coding Černý, Jan ISIS:116092 ths Škrabálek, Jaroslav opn Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky kn20010709399 dgg https://insis.vse.cz/zp/93476/podrobnosti VŠKP v InSIS https://aleph.vse.cz/neverejny_soubor_prace.pdf Hlavní práce https://insis.vse.cz/zp/93476/posudek/vedouci Hodnocení vedoucího https://insis.vse.cz/zp/93476/posudek/oponent/89592 Oponentura https://aleph.vse.cz/neverejny_soubor_prace.pdf Přiloha k práci https://aleph.vse.cz/neverejny_soubor_prace.pdf Přiloha k práci https://aleph.vse.cz/neverejny_soubor_prace.pdf Přiloha k práci https://insis.vse.cz/zp/93476/podrobnosti dc:identifier NEPOSILAT VSKP vse93476 260312 93476