In the rapidly evolving business landscape, small and medium-sized enterprises (SMEs) are increasingly seeking to enhance operational efficiency through process optimization and automation. The advent of no-code and low-code platforms has facilitated access to business process automation tools, allowing non-technical users to design and implement workflows. However, these tools often lack the ability to provide domain-specific guidance tailored to the unique needs and language of individual customers. This gap hinders effective process automation, placing a significant burden on SMEs to interpret complex strategies without a clear understanding of their operational context. The challenge is ensuring that automation aligns with each company’s unique domain and strategic objectives, making the automation journey effective and comprehensible for all stakeholders. This project aims to develop a support system that empowers customers to better understand and make informed decisions about process automation. The system will facilitate sophisticated, context-aware conversations tailored to the customer's domain, providing intuitive interactions and domain-specific examples. This will help customers navigate critical decision-making junctures, ensuring that their choices align with their strategic goals and enhance operational efficiency. This approach is crucial for enabling customers to fully leverage the benefits of process automation in their unique business environments.

The proof-of-concept prototype must access domain-specific information, ask insightful questions to accurately capture customer needs, explain complex concepts clearly, and generate formal process specifications from decisions made during interactions. This study will test the feasibility of an assistant that uses a generative large language model (LLM), retrieval augmented generation (RAG), and state machine-based dialog management to meet these requirements.

The research questions include:

• How can we model conversational states and transitions for business process automation requirements engineering?
• How can prompts be augmented with domain-specific information to tailor responses to the customer's context?
• How can UML-based specifications be extracted from conversations via the language model?

The project aims to deliver a codebase demonstrating the feasibility of these functionalities and provide answers to the stated research questions.

Our research develops advanced Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and state-based dialogue frameworks to enhance business process automation (BPA). This integration aims to bridge the gaps of existing BPA tools, which often rely on static data and lack personalization and adaptability to dynamic changes. Our solution uses continuous interaction and learning from Generative LLMs, enabling personalized, real-time adaptability and industry specific information retrieval. By integrating RAG and dialogue frameworks, the system dynamically adjusts to user behaviors and business contexts, enhancing decision-making and optimizing process designs. This approach promises improved efficiency, accuracy, and user experience, potentially transforming AI-assisted decision-making across industries. Further research will focus on integration with existing platforms and addressing security concerns, aiming for broader adoption.

The iDPARC innovation project aims to enhance its business model's scalability by automating the client requirements assessment phase, critical for client business automation projects. Over eight years, iDPARC has successfully marketed digital process automation software to SMEs but struggles with scalability due to the labor-intensive requirements assessment, involving senior engineers and clients, which prolongs project cycles. To address this, an AI assistant will automate the assessment, generating structured requirements in meta-language and Unified Modeling Language (UML), integrating directly into iDPARC’s software to improve scalability. A feasibility study will validate the AI assistant’s functionalities, leading to an Innosuisse project to develop a minimum viable product (MVP).

The project will progress in phases:

• Collaborative evaluation of MVP requirements with a research partner, involving iterative design and client feedback.
• Prototype refinement through design-test-learn cycles and user tests for product-market fit.
• Market testing and final adjustments of the MVP, fine-tuning the business model and documentation before project completion.

This feasibility study will provide essential learning aids to define the MVP's core features, supported by a comprehensive Innosuisse project to enhance iDPARC’s efficiency and market reach.