AI Decision Assistant ChatBot for Software Release Planning and Optimized Resource Allocation
Ehab Ebrahim, Mazen Sayed, Mahmoud Youssef, Hisham Essam, Salma Abd El-Fattah, Dina Ashraf, Omar Magdy, Reem ElAdawi · 2023
This paper presents Alfred, a web-based artificial intelligence (AI) chatbot powered by the Rasa framework and designed to facilitate Agile software release planning by using two machine learning (ML) models to estimate the time for each task selected by the project managers and recommend the optimal resource to be assigned to each task. Extreme gradient boosting (XGBoost) is used for time estimation, and the model’s performance evaluated using accuracy, adjusted coefficient of determination (R2), and fitting a line between the estimation and actual values. The results obtained were 82% accuracy and 98% adjusted R2 score. K-nearest neighbors (K-NN) is used for resource recommendation in a one-shot learning manner. The application integrates a dashboard with useful visualizations and an interactive smart assistant AI-based chatbot, providing project managers with a tool to make more informed decisions. Additionally, an approach is used to indicate the model’s confidence in tasks estimation of given tasks based on the amount of available data showing the history in the system to give project managers more insights. The approach uses a keywords frequency indication algorithm based on the summary tokens to group the task estimates into one of three categories of confidence: high, medium, or low.