A WEB TOOL FOR NEW PRODUCT DEVELOPMENT

The development of new products in industrial contexts has traditionally relied on fragmented approaches where design of experiments, machine learning, and knowledge discovery operate as isolated methodologies. This fragmentation creates inefficiencies in innovation workflows. The thesis addresses this challenge by designing and developing an integrated web-based platform that unifies these components into a cohesive system for industrial product development. The platform integrates two main components within a structured workflow. First, an AI-powered assistant leveraging large language models with retrieval-augmented generation facilitates knowledge discovery, helping users identify promising experimental directions from curated technical databases. Second, the system implements an alternative version of ALPERC (Active Learning for Physical Experiments based on nonparametric Ranking and Clustering), an established iterative framework that combines efficient experimental design with machine learning-based optimization, while also providing actionable insights. The ALPERC algorithm's effectiveness is demonstrated through a case study in materials science involving the prediction of critical transformation temperatures in amorphous metallic alloys, showing that the framework achieves comparable predictive accuracy while requiring 85% less experimental data than traditional full-dataset approaches. The AI assistant was independently evaluated on a chemical industry use case, achieving high factual accuracy (mean faithfulness score of 4.54/5.0) using Claude 3.5 Haiku with metadata-enhanced retrieval. The developed system provides R&D professionals with a complete platform that streamlines the product development process from initial exploration through iterative optimization, with potential applications across diverse industrial domains.