Predictive Insights for your Software Tasks
DevDucky helps your product owners to make the right planning decisions
Today you need to gather information from your developers, asking them questions like how complex is this task?
How long do they think it will take them? Sometimes you don't even know who to ask these questions, and most of the time you hear guesses from the developers.
Get fast and accurate information to make the right planning decisions
instead of having meetings and guesses, you can use our AI model to gain accurate information about a task for a specific developer. Our models take the task information: Text description, the developer, who made the task, and more. and outputs Insights like the estimated duration of the task for the given developer, how complex it will be and more.
Our models take the task information: Text description, the developer, who made the task and more, and outputs Insights like the estimated duration of the task for the given developer, how complex it will be, who are the developers who are best to go and ask questions regarding the task and more.
We used the model on Apache's open source projects, you can test it on our live demo website
How does DevDucky works?
The Predictive Insights is made of 2 main building blocks.
The 1st block is a model that was trained to connect the business aspect to the code. given a task, we apply the model to the entire code repository and identify what areas of the code require modifications and how.
This ability was achieved by utilizing deep learning algorithms and rule-based classic methods.
The 2nd block uses the output of the 1st block and extracts the insights from the relevant code and issue data.
it analyzes the complexity of the task as a function of both the complexity of the code and the description of the issue and uses multiple methods to refine it into a single value.
A deep learning regression model is used to estimate the task duration for the given developer. the model learns a unique vector representation for each developer and issue creator, in addition, it takes as an input the output of the 1st block and features from the issue like the text and issue type. the model achieves 88% accuracy on the test set.
Developers who worked on a similar task before or who knows the area of the code will be suggested as Expert Developers
Related Components and Effected Endpoints are extracted from the code that is to be modified for the task.
The confidence metric is aggregated from the other metrics results.
Reduce planning errors by giving more accurate information
Your developers can focus on creating values
Prevent errors by finding the right developer for the task
faster feature launching by correct prioretizing