2026 ASEE Annual Conference & Exposition

In this work-in-progress paper, we report findings from the second year of implementation of the WaterSoftHack data science workshop, a virtual three-week training program situated in water science that culminates in a team-based hackathon. This work is part of an NSF CyberTraining project utilizing a design-based research approach to develop data science knowledge and skills in the water sciences. Building on formative findings from Year 1, the Year 2 iteration focused on refining the program structure while maintaining the core elements: intensive training, applied project work, and collaborative hackathon experience.

Participants were eighteen Fellows, including graduate students, postdoctoral researchers, and faculty members seeking professional development in data science. Fellows participated virtually, collaborating with teams on final hackathon projects. Data collection included pre- and post-workshop surveys and in-depth semi-structured interviews, with all five returning Fellows from Year 1 completing interviews. Given the small matched cohort, we triangulated quantitative and qualitative analyses using the Kirkpatrick and CIPP evaluation frameworks to provide a comprehensive understanding of participant experiences.

Our analysis reveals several significant findings. First, the program successfully boosted participant self-efficacy and facilitated the acquisition of advanced computational skills. Matched cohorts demonstrated consistent gains in self-rated proficiency across core competencies, including cyberinfrastructure prototyping, machine learning programming, and cloud computing. Participants explicitly credited hands-on practice and accessible instructional materials, such as Google Colab notebooks, for their progress. Fellows reported new proficiency with advanced models, including Transformers and LSTMs, and increased confidence to apply these methods in their research.

Most significantly, we observed a high rate of immediate knowledge transfer and tangible research impact. Interviews documented direct application of workshop learning to dissertation projects and coursework, sustained professional networking among participants, and active plans to publish hackathon work. Multiple teams identified publication targets and established timelines for manuscript drafting, demonstrating a direct translation of workshop activities into valuable research outputs.

Despite these successes, persistent and significant challenges emerged that mirrored and intensified Year 1 patterns. The compressed three-week timeline was consistently cited as creating an overwhelming pace and steep learning curve. Team dynamics challenges were more pronounced in Year 2, with uneven prior knowledge among team members leading to interpersonal conflict. This friction was reportedly compounded by perceived gaps in faculty support during conflicts, highlighting a critical need for structured socio-technical support mechanisms.

Participants provided actionable recommendations for future iterations, including: implementing a "Week Zero" scaffolding period with targeted readings on specific machine learning models; earlier team formation with proactive attention to skill-level matching; more guided code walkthroughs explaining parameter choices; streamlining Week 1 theory to focus on models used in the hackathon; and establishing clear protocols for conflict resolution. Several Fellows also suggested exploring a three- to four-week program arc to reduce the intensity of the learning curve.

These findings offer crucial insights for educators and program designers developing intensive, team-based computational training for the STEM research community. Implications for cybertraining design include centering applied, domain-authentic tasks; providing just-in-time theoretical scaffolds; and structuring collaboration supports that reduce skill-mismatch friction.