We combine deep engineering with generative AI. We start by creating a highly detailed graph that represents each element of your software (modules, functions, procedures, classes, etc.) and database (tables, stored procedures, functions, views, etc.) and all the relationships between those elements (functions that call procedures, global objects in use, procedures that read or write tables, etc.).
From there, we have a pipeline of more than 60 steps that create backend APIs and then frontend forms, connecting the forms to the APIs to reproduce the functionality. These steps are deterministic in what they do and how they do it, but they rely on generative AI to produce each piece of code. The graph provides exactly the right context for each small step.
For example, to create an endpoint for a newly created API, the system will gather from the graph the exact methods, procedures, and tables with AI-generated descriptions and their relationships, to give the LLM the proper context it needs, no more, no less. Subsequently, additional agents will review the code, correct or improve it if necessary, ensure that the client's standards are met, and verify that the endpoint fits perfectly into the big picture.