Adding AI Capabilities to Glyue
Alexander Emili
Sandbox Banking is proud to announce the integration of OpenAI’s GPT-3.5 model into Glyue to provide code suggestions! This new functionality was part of our v4.0.2.9 release earlier this month, and more AI-based enhancements are planned throughout 2023.
Like much of the technology world, Sandbox Banking’s product and engineering teams have been excited by the recent large deep learning models recently released by OpenAI and others. We believe the coming generations of models will deeply transform the software integration landscape and software development in general.
GitHub Copilot – a trained AI assistant to developers that effectively acts as a virtual pair programmer – is an early example of the technology’s usefulness in our industry. In a recent controlled experiment performed by GitHub, a higher proportion of professional developers using GitHub CoPilot were able to complete a web server coding task (78%) than a similar group of developers without the tool (70%). Furthermore, the developers using Copilot that completed the task were able to perform the work 55% faster! GitHub surveys also indicate that 60-75% of Copilot users feel more fulfilled at their job due to the tool. More information is available in this research article from Github’s blog.
GitHub isn’t an unbiased source of information regarding the impact and efficacy of Copilot, but our team’s own anecdotal experience with the tool suggests that its a massive boon to engineering productivity. Given our aspiration to provide the universe’s best banking integration builder, the decision to build Copilot-like code generation into Glyue was a no-brainer.
Check out this video to see Glyue’s new AI code generation functionality:
The code suggestions dialog leverages the GPT 3.5 model to provide a suggestion based on a user prompt. The tool considers the surrounding code/logic context, and with a description of the desired output can generate a potential solutions to the task at hand. Usually a comment below the code helps the model best determine a relevant response, but the prompt can handle pure code inputs as well. Users can request multiple suggestions based on the same prompt and choose the best solution from a variety of choices. With the insert suggestion button, the code snippet will be inserted into the open editor.
Future Glyue AI Features
Some future potential Glyue AI features are code block explanations, a more seamless integration of the code suggestions, and AI assistance with creating whole integrations. Our team is also investigating the documentation generation capabilities of these models. Deep learning classification models also have important implications for production support automation and improved issue/error handling.
This is an incredibly exciting time within the artificial intelligence space as these models become further democratized and improved – this technology is poised to massively accelerate the innovation and digital transformation journeys of Sandbox Banking’s customers.
