ArtificialCast provides a C# utility for type-safe object transformation and generation powered by local Large Language Models (LLMs). It aims to replace manual mapping code with LLM-driven inference, allowing seamless conversion between disparate data structures based solely on type definitions and prompts. This is targeted at developers exploring the implications of "virtual software" where behavior is inferred rather than explicitly coded.

How It Works

The core mechanism involves serializing input objects, generating a JSON schema for the target type, and then prompting a local LLM (configured via ArtificialCast.Model) to perform the transformation. This approach leverages the LLM's reasoning capabilities to infer missing values and adapt between types without explicit mapping rules, offering a "zero-config" experience for transformations.

Quick Start & Requirements

• Install: Clone the repository, add a project reference, and run dotnet restore.
• Prerequisites: .NET 9 SDK or later, Ollama installed and running, and a local LLM model (e.g., gemma3:4b via ollama pull gemma3:4b).
• Setup: Requires cloning the repo, adding project references, and pulling an LLM model.
• Docs: README.md serves as primary documentation.

Highlighted Details

• Enables "virtual software" where behavior is inferred by LLMs from types and prompts.
• Demonstrates a "failure mode of success" where outputs appear correct but lack semantic accuracy.
• Offers functions like AC<TIn, TOut> (transform), AF<T> (factory), AM<T1, T2, TOut> (merge), AS<TIn, TOut1, TOut2> (split), and AQ<TOut> (query).
• Operates locally without API keys or cloud dependencies.

Maintenance & Community

The project is maintained by Zorokee. The README indicates it's a demonstration of AI engineering trends and is not intended for production use.

Licensing & Compatibility

• Source Code License: "NOT FOR USE LICENSE" - strictly prohibits use in production, testing, demo, or internal tooling. Permitted for private study, educational purposes, and academic citation.
• Documentation License: CC BY 4.0 - permits sharing and adaptation for academic/educational use with attribution.
• Compatibility: Not compatible with any deployed system or commercial use.

Limitations & Caveats

The project is explicitly designed as a demonstration and is fundamentally unsafe for any real-world application. Its outputs are described as "plausible lies" and "hallucinations," with failure modes that mimic success, making it unreliable for critical tasks.