Tapis AI Assistant: Increasing Research Impact through Traceable and Evaluated Agentic Tools
Year: 2026
Type: conference-paper
Abstract: AI Agents are rapidly gaining popularity by significantly reducing the “time-to-science” while performing complex scientific tasks, which previously required hours or days of human effort. While modern cyberinfrastructures like Tapis provide powerful, multi-service capabilities for managing distributed research, understanding the services and integrating them into scientific workflows can be a steep learning curve. For new users to navigate quickly and maximize the impact of such robust systems, research software engineers (RSEs) are building specialized agentic frameworks using Model Context Protocol (MCP) tools. By engineering these standardized interfaces, RSEs can lower the entry barrier and accelerate domain-specific research.In this paper, we present a case study of the Tapis AI assistant, a system built to address this bottleneck by providing an intuitive, agentic interface that extends the capabilities of the Tapis framework. The users of this system can quickly understand Tapis services and integrate them into their scientific workflows. Most importantly, this framework addresses the AI ``black-box" observability and tracing issue by integrating traceability via Arize Phoenix. Furthermore, the agent evaluation discussed in the paper ensures the reliability of agentic systems for high-stakes research. With this study, we demonstrate that by instrumenting and evaluating AI agents to incorporate transparency and reliability, RSEs can build scalable and dependable research pipelines.
Authors: Padhy, Smruti, Jamthe, Anagha, Stubbs, Joe
Institution: The University of Texas at Austin
Language: English
Subfield: Information Systems and Management
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7214335733
