Research tasks during the conceptual phase of product development are crucial for innovation, yet they are time-intensive and susceptible to error. Prevailing AI tools, which typically rely on a agent-centered architecture to orchestrate the research process, are constrained by the finite context window of language models. This limitation hinders the execution of comprehensive research required in industrial environments. This thesis addresses this challenge by developing DataMind, an application that implements a novel architecture based on programmatic orchestration and a structured database. In this approach, an AI agent conversationally deconstructs complex research goals into granular, interdependent tasks and then a programmatically orchestrated system manages the execution of these tasks, storing all findings in a structured database, which enables greater scalability and research scope compared to existing solutions.The system's efficacy was validated through a case study involving a market analysis of unmanned ground vehicles (UGVs), an industrial task provided by the company DOK-ING. The results demonstrate that DataMind significantly outperforms both manual methods and selected commercial AI tools. Specifically, the application identified an order of magnitude more relevant UGV platforms and gathered more detailed information, delivering the output as an interactive database rather than a static report. This thesis therefore presents and validates a novel architectural approach for AI-based tools that, by leveraging programmatic orchestration and structured storage, overcomes the limitations inherent in single-agent-centric architectures.
