Building a research-software catalog with a coding agent: from hackathon prototype to public deployment

Generative AI and coding agents are accelerating research software development, necessitating improved methods for software discovery and maintenance. A three-day hackathon prototype for a repository catalog was developed, highlighting critical engineering requirements for public deployment, including adversarial review and data quality. The lessons from this prototype are being explored for application to larger, human-curated portals using retrieval agents combining various data sources and AI capabilities.

Why it matters

The emergence of generative AI in software development introduces both efficiency gains and new operational complexities, particularly around managing and discovering software assets. This development signals a shift towards AI-assisted tool creation and necessitates proactive strategies for data governance, quality assurance, and scalable infrastructure to support increased software output.

Key insights

• Generative AI and coding agents are noted for their potential to accelerate research software development.
• This acceleration increases the need for efficient software discovery and maintenance mechanisms.
• A research software catalog prototype was developed within a three-day hackathon timeframe.
• Making this prototype suitable for public deployment required addressing engineering challenges such as adversarial review, data-quality checks, browser-level validation, and publication safeguards.
• The methodology is being investigated for transferability to larger, human-curated portals like MateriApps, utilizing retrieval agents with curated metadata, external documentation, vector search, and local language model generation.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.04711