OSSPREY: AI-Driven Forecasting and Intervention for OSS Project Sustainability

---

# Description

This paper presents OSSPREY, an AI-powered platform designed to forecast and support the sustainability of GitHub-hosted open source projects. OSSPREY addresses the widespread challenge that nearly 90% of OSS projects stagnate or are abandoned, leaving critical software infrastructure vulnerable. Unlike existing tools such as Apache Clutch or APEX, which are either foundation-specific or lack actionable guidance, OSSPREY integrates three capabilities: real-time monitoring of project activity, month-by-month sustainability forecasting, and evidence-based actionable recommendations. Its architecture combines a Rust-based scraper, socio-technical network generator, transformer-based forecasting module, and a ReACT recommender system that maps project underperformance to research-backed interventions. A modular dashboard visualizes project health trajectories, socio-technical networks, and tailored interventions, enabling maintainers to monitor, anticipate decline, and apply strategies supported by empirical studies. OSSPREY thus links sustainability research with actionable decision-making at scale.

# Findings

We find that OSSPREY provides maintainers with the ability to forecast project sustainability with high accuracy and to link downturns to concrete, evidence-based interventions. We show that by scraping GitHub repositories in real time and generating socio-technical networks, OSSPREY can compute rich feature sets that allow a transformer-based model to predict sustainability probabilities month by month. We demonstrate that the ReACT recommender identifies underperforming features in these socio-technical metrics and provides tailored interventions ranked by importance and supported with references to peer-reviewed studies. We further find that the dashboard integrates these forecasts and recommendations into an accessible interface, allowing maintainers to visualize alternative future trajectories and take timely corrective action. We conclude that OSSPREY bridges the gap between sustainability research and practice by offering an integrated framework for monitoring, forecasting, and intervention, while also noting current limitations such as reliance on fixed feature sets, static mappings of interventions, scalability constraints for very large repositories, and lack of ecosystem-level dependency modeling. We argue that extending OSSPREY to support adaptive modeling, context-aware recommendations, and ecosystem-wide analysis will further strengthen its role as a sustainability analytics platform for open source software.