SmartSim: A Curriculum-Centric Conversational Agent for Employee Training
ales and service training often lacks personalized, scalable practice tools, leading to inconsistent skill application on the job. In our experience working with sales and service training teams, tools they use often lack personalized, scalable practices, leading to inconsistent skill application on the job. Corporate training seldom provides structured, realistic practice, resulting in low retention and limited behavior change. We built a curriculum-centric, multi-agent framework for role-play simulations (could be adapted for text and voice) that deliver sequenced, educationally grounded training conversations for sales employees. The framework automates delivering curriculum of scenarios by coordinating several agents: the Orchestrator manages flow, the Curriculum Manager sequences role-play scenarios with embedded learning objectives and rubrics, the Conversation Agent enacts realistic dialogues, the Data Agent tracks progress, Telemetry logs outcomes, and the Guard enforces safety. We designed a workflow prototype in n8n that simulates the behavior of a phone-based conversational agent. For this paper, the system is demonstrated through a chat-based interface that reproduces the curriculum sequencing and orchestration logic, rather than a production telephony deployment. Pilot learners told us that sequenced curricula improved practice consistency and in turn confidence. Learners appreciated how realistic practice was and specific feedback via Telemetry, although challenges still remain around sustaining engagement and avoiding repetitive feedback. Through these findings we canconclude that a curriculum-centric, multi-agent role play simulation can improve learning outcomes in employee training, bridging the gap between adaptive e-learning and real-world application. Our objective is to examine whether curriculum-guided AI role play can improve soft skills when compared to unstructured. This paper contributes a curriculum-centric framework, functional prototype and findings from a formative pilot with three sales hires.
Authors
- Anke John
- Tejas Sarvankar
Institutions
- Carnegie Mellon University (US)
Publication Details
- Journal
- The American Journal of Engineering And Technology
- Published
- 2025-10-10
- DOI
- https://doi.org/10.37547/tajet/volume07issue10-05
- Primary Topic
- AI in Service Interactions
- Type
- article
- Field-Weighted Citation Impact
- 0.00