AI coding engine generates code based on the input configuration and specified rules. It is less error-prone. This engine helps save the time taken to resolve the compilation error. It can produce code every second of the day. It can learn the code from a repository or different code base. It can improve code generation efficiency by learning design and coding best practices. AI coding engine based on Generative AI provides better ROI and benefits by improving efficiency and cutting down defects in the system. Generative AI frameworks can produce code in the different tech stack that consists of UI/Rest Services/Data Access frameworks. Code generation and code documentation will become simpler and faster.

Initially, this coding engine is about an AI Bot that can generate code in PHP, Ruby on Rails, Django, and Java. This coding engine based on Generative AI can develop code in python and Django or any other frameworks. Typically we provide details regarding the software and how it works. This process takes effort, which is equal to writing the code. It needs basic info about the entities involved in the code. It can train itself by reading zillions of human-written code. It can read from a repository and generate code based on the components. It is helpful to create samples of code and unit tests for software frameworks. Data generation for testing is another feature of this Coding engine. It can take input as rules, tools, definitions, protocols, database support, or other software support. It can be enhanced to support developed software API for code generation. It is based on neural sketch learning. Neural sketch learning is related to an ANN trained to identify code-level patterns from different software.

A neural sketch-based mobile application framework mixes neural and combinatorial methods for displaying and rendering native, hybrid, and web-based mobile applications. It is applicable for iOS, Android, and Windows operating system-based devices and different programming language SDKs supported by the devices.

Supervised learning is the method used for training the mobile application framework. The training input has a set of applications. They are configured based on views, entities, and actions. The metadata provided by the application configuration helps the framework learn the application’s semantics. The application has data types, methods, control flow, exceptions, and type-based entities.

The neural sketch-based framework identifies the patterns. It helps in rendering and displaying the application. The application needs to satisfy the constraints specified in the configuration file. The neural sketch framework is rooted in a mix of neural learning and combinatorial search. The source code in the framework can create tree-based syntactic models and sketches.

A sketch is related to a class, entity, and method. The methods have arguments, return values, and control flows. The architectural approach is based on the metadata-driven architecture pattern. The metadata is stored and modeled using the object repository pattern. The code reads the metadata and renders the mobile application assuming microservices-based architecture. The architecture has a service layer, mobile services client, object to the relational persistence layer, data access layer, and database layers. These layers are implemented using the API and architectural patterns. The metadata attributes are configured in the input for different layers. The application is configurable and extensible with events, delegates, templates, services, and classes.

 The code in the framework uses the metadata for handling different layers in the language SDK and platform. The documentation is configured, and it is then used for application rendering and displaying. The user interface can be complex and will need layouts and additional configuration for different UI controls. The complex business logic will be passed on through custom modules and classes. The framework has capabilities to handle upgrades, versions, and maintenance-related features.

The neural sketch approach is based on the latest technology and can be upgraded to any new technology stack by changing the application framework code stack in different layers. The database layer persists the data. It also handles the current state. The future scheduled state modifications can be handled by the database layer. The persistence layer has features to store and retrieve data from various data sources. The retrieval of the data is done by using different adapters. The business logic and the rules layer handle the complex changes across the business entities and the entity groups. The service layer handles the updates and manages the group of entities. The user interface layer has features related to customization, configurability, and extensibility of the attributes and entities.

Different organization use metadata-driven approaches for creating mobile application framework-based products. (e.g., Oracle). Metadata is used to create the application by dynamic patterns and generation of the presentation layer. The other layers are modelled in the input configuration files. The modelling is based on generic attribute-based entity patterns.

The application framework helps in upgrades, customizations, maintenance, and other challenges in the software engineering lifecycle. The provided user interface handles the customizations. These customizations do not require any code changes. The custom modules can be coded which can be loaded by the application framework. 

The application user interface will be typically simple and easy to use. The application will render in different languages, formats, and organization/layouts based on the user’s location and setup. The application will have support for globalization, which includes internationalization and localization. The application will have rapid and flexible searching and reporting. Collaboration and Data Sharing will be part of the application features. The application will support data sharing with different non-product users, inside and outside of the system. 

[1] Neural sketch learning for conditional program generation V Murali, L Qi, S Chaudhuri, C Jermaine [2] Neural attribute machines for program generation M Amodio, S Chaudhuri, T Reps [3] Deepcoder: Learning to write programs M Balog, AL Gaunt, M Brockschmidt, S Nowozin