Pablo Fernández

Cesar Garcia, Alejandro Guerrero, Joshua Zeitsoff et al.

Agile software teams are expected to follow a number of specific Team Practices (TPs) during each iteration, such as estimating the effort ("points") required to complete user stories and coordinating the management of the codebase with the delivery of features. For software engineering instructors trying to teach such TPs to student teams, manually auditing teams if teams are following the TPs and improving over time is tedious, time-consuming and error-prone. It is even more difficult when those TPs involve two or more tools. For example, starting work on a feature in a project-management tool such as Pivotal Tracker should usually be followed relatively quickly by the creation of a feature branch on GitHub. Merging a feature branch on GitHub should usually be followed relatively quickly by deploying the new feature to a staging server for customer feedback. Few systems are designed specifically to audit such TPs, and existing ones, as far as we know, are limited to a single specific tool. We present Bluejay, an open-source extensible platform that uses the APIs of multiple tools to collect raw data, synthesize it into TP measurements, and present dashboards to audit the TPs. A key insight in Bluejay's design is that TPs can be expressed in terminology similar to that used for modeling and auditing Service Level Agreement (SLA) compliance. Bluejay therefore builds on mature tools used in that ecosystem and adapts them for describing, auditing, and reporting on TPs. Bluejay currently consumes data from five different widely-used development tools, and can be customized by connecting it to any service with a REST API. Video showcase available at governify.io/showcase/bluejay

Pablo Ariño Fernández

The job shop scheduling problem is an NP-hard combinatorial optimization problem relevant to manufacturing and timetabling. Traditional approaches use priority dispatching rules based on simple heuristics. Recent work has attempted to replace these with deep learning models, particularly graph neural networks (GNNs), that learn to assign priorities from data. However, training such models requires customizing numerous factors: graph representation, node features, action space, and reward functions. The lack of modular libraries for experimentation makes this research time-consuming. This work introduces JobShopLib, a modular library that allows customizing these factors and creating new components with its reinforcement learning environment. We trained several dispatchers through imitation learning to demonstrate the environment's utility. One model outperformed various graph-based dispatchers using only individual operation features, highlighting the importance of feature customization. Our GNN model achieved near state-of-the-art results on large-scale problems. These results suggest significant room for improvement in developing such models. JobShopLib provides the necessary tools for future experimentation.