API World + AI DevWorld

We are coming closer to a reality where computers can generate new programs that solve new problems based on limited human input or natural language. Program synthesis is attempting to capture human intent using high-level descriptions and perhaps a conversational interaction to generate working code. Currently, techniques for constructing programs in general purpose programming languages are not in reach. However, if we restrict the domain of the problem and the type of language we are targeting we are seeing some useful results. The general area of data, API, and workflow automation offers such a restricted problem domain, and this is the domain for which have been developing program synthesis techniques at SnapLogic.

Program synthesis requires a means for humans to describe the outcome of the desired synthesized program. This can be achieved in a variety of ways and while some are aspirational, like arbitrary natural language, the first useful program synthesizers will be ones that generate partially complete programs that require human refinement. The current generation of machine learning models and deep learning frameworks have great potential to make program synthesis more robust and capable.

In this session, Professor Greg Benson will explain the general problem of program synthesis and provide an overview of recent research results as well as current challenges for achieving full and partial synthesis. As a concrete example, Greg will explain the challenges we have faced at Snaplogic in building a synthesizer for the SnapLogic visual programming language, which is used for data, API, and workflow automation. Attendees will come away with a better understanding of the state of the art in program synthesis and a view to the future as more platforms incorporate this technology.