AI Coder- Chartreuse

Please check out new branch added - Chartreuse
https://github.com/bhagvank/AI-Coder/blob/Chartreuse/README.md

How It Works

Neural Sketch Learning

Code generation is driven by a syntax and semantics specification supplied as input. The framework is trained by being shown applications that are already configured within it (a supervised learning approach). Training input consists of a set of applications configured in terms of their views, entities, and actions; metadata from these configurations teaches the framework the application's semantics — its data types, methods, control flow, exceptions, and entity types.

The neural sketch framework combines neural learning and combinatorial search to identify patterns and render applications, while enforcing the constraints defined in the configuration file. Internally, the source code can build tree-based syntactic models ("sketches") of a program.

A sketch corresponds to a class, entity, or method — including its arguments, return values, and control flow.

Application Framework

The framework follows a metadata-driven architecture, with metadata stored and modeled using the object repository pattern. Metadata is read to render applications assuming a microservices-based architecture with the following layers:

• Service layer
• Services client (mobile)
• Persistence layer
• Data access layer
• Database layer Each layer is implemented using language SDK– and platform-specific patterns and practices, with metadata attributes configured per layer. Applications built with the framework are configurable and extensible through events, delegates, templates, services, and classes, and support customization without code changes (with custom modules loadable when code changes are needed).

This approach targets native, hybrid, and web-based mobile applications, and can be used across iOS, Android, and Windows, along with their respective programming language SDKs. Because the approach is not tied to a specific technology stack, it can be upgraded by changing the framework's implementation in the relevant layer.

Application Architecture

The application architecture is designed to be extended with commercial packages and software, using only the functionality required, and includes:

• Security & licensing enforcement
• Auditing of inserts, edits, and deletions
• Multi-tiered, entity-context sensitive security for data and actions across user roles and groups
• Admin-level metadata management through screens, plus developer-level configuration of metadata and programmatic update