Determinant is an early-stage experiment that proposes Auditable Application Language (AAL) as a deterministic layer between probabilistic AI generation and executable software. The system allows an LLM to generate AAL from natural language requirements, which humans then review for business logic correctness before a traditional compiler generates TypeScript code.
- AAL focuses on two primary concepts: Object (describing application state) and Flow (describing behavior), avoiding implementation details like classes or methods.
- The compilation pipeline converts accepted AAL into an AST, performs semantic checks and binding, and outputs deterministic TypeScript for Node.js without further LLM involvement.
- Current features include typed fields, conditions, explicit failures, calculations, state changes, flow composition, money types, and support for English and Chinese dialects.
- The project aims to reduce the scope of AI uncertainty by letting humans audit a smaller, executable description rather than reviewing thousands of lines of generated code.
The author seeks feedback on whether this review boundary is practically useful and if AAL can remain readable while covering sufficient application behavior.