Document intelligence is undergoing a significant paradigm shift, transitioning from tradi- tional spatial and structural analysis to generative artificial intelligence and agentic workflows within enterprise environments. Driven by an industrial internship, this work bridges advanced theoretical foundations with practical enterprise automation. The study examines the inte- gration of generative prompting within automated business workflows, specifically using Mi- crosoft Power Platform and AI Builder. Grounded in the theoretical principles of Large Lan- guage Models, including transformer architectures and latent space representations, the method- ology formalizes core prompt engineering techniques such as few-shot learning, reasoning, and prompt chaining, to optimize document parsing and mitigate common anti-patterns. Look- ing forward, the research extends to geometric extraction techniques and provides a rigorous performance comparison between specialized document intelligence frameworks and genera- tive AI solutions. Ultimately, these methodologies are deployed in a real-world application for automated order processing and sales agent analysis. This work demonstrates how systematic prompt engineerifffgng serves as a critical mechanism for scalable, data-driven decision-making and intelligent automation in contemporary corporate operations.

Document intelligence is undergoing a significant paradigm shift, transitioning from tradi- tional spatial and structural analysis to generative artificial intelligence and agentic workflows within enterprise environments. Driven by an industrial internship, this work bridges advanced theoretical foundations with practical enterprise automation. The study examines the inte- gration of generative prompting within automated business workflows, specifically using Mi- crosoft Power Platform and AI Builder. Grounded in the theoretical principles of Large Lan- guage Models, including transformer architectures and latent space representations, the method- ology formalizes core prompt engineering techniques such as few-shot learning, reasoning, and prompt chaining, to optimize document parsing and mitigate common anti-patterns. Look- ing forward, the research extends to geometric extraction techniques and provides a rigorous performance comparison between specialized document intelligence frameworks and genera- tive AI solutions. Ultimately, these methodologies are deployed in a real-world application for automated order processing and sales agent analysis. This work demonstrates how systematic prompt engineerifffgng serves as a critical mechanism for scalable, data-driven decision-making and intelligent automation in contemporary corporate operations.

Deploying an Intelligent Sales Agent for Autonomous Order Processing in Enterprise Environments

CARTURAN, MARCO AUGUSTO
2025/2026

Abstract

Document intelligence is undergoing a significant paradigm shift, transitioning from tradi- tional spatial and structural analysis to generative artificial intelligence and agentic workflows within enterprise environments. Driven by an industrial internship, this work bridges advanced theoretical foundations with practical enterprise automation. The study examines the inte- gration of generative prompting within automated business workflows, specifically using Mi- crosoft Power Platform and AI Builder. Grounded in the theoretical principles of Large Lan- guage Models, including transformer architectures and latent space representations, the method- ology formalizes core prompt engineering techniques such as few-shot learning, reasoning, and prompt chaining, to optimize document parsing and mitigate common anti-patterns. Look- ing forward, the research extends to geometric extraction techniques and provides a rigorous performance comparison between specialized document intelligence frameworks and genera- tive AI solutions. Ultimately, these methodologies are deployed in a real-world application for automated order processing and sales agent analysis. This work demonstrates how systematic prompt engineerifffgng serves as a critical mechanism for scalable, data-driven decision-making and intelligent automation in contemporary corporate operations.
2025
Deploying an Intelligent Sales Agent for Autonomous Order Processing in Enterprise Environments
Document intelligence is undergoing a significant paradigm shift, transitioning from tradi- tional spatial and structural analysis to generative artificial intelligence and agentic workflows within enterprise environments. Driven by an industrial internship, this work bridges advanced theoretical foundations with practical enterprise automation. The study examines the inte- gration of generative prompting within automated business workflows, specifically using Mi- crosoft Power Platform and AI Builder. Grounded in the theoretical principles of Large Lan- guage Models, including transformer architectures and latent space representations, the method- ology formalizes core prompt engineering techniques such as few-shot learning, reasoning, and prompt chaining, to optimize document parsing and mitigate common anti-patterns. Look- ing forward, the research extends to geometric extraction techniques and provides a rigorous performance comparison between specialized document intelligence frameworks and genera- tive AI solutions. Ultimately, these methodologies are deployed in a real-world application for automated order processing and sales agent analysis. This work demonstrates how systematic prompt engineerifffgng serves as a critical mechanism for scalable, data-driven decision-making and intelligent automation in contemporary corporate operations.
Generative AI
Workflow Automation
LLM
Document Parsing
Sales Agent
File in questo prodotto:
File Dimensione Formato  
Carturan_MarcoAugusto.pdf

Accesso riservato

Dimensione 1.32 MB
Formato Adobe PDF
1.32 MB Adobe PDF

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/115826