ICSE2027demo-Req2Web

Early-stage software requirements are often incomplete, leaving interface structure, interaction states, failure handling, and acceptance conditions unspecified. Generating a runnable prototype therefore requires not only filling these gaps, but also retaining evidence associated with design decisions and checking whether the resulting behavior satisfies the original requirements. We present Req2Web, a retrieval-augmented tool for generating runnable and traceable web prototypes from informal natural language requirements. Req2Web structures requirements into use cases and constraints, retrieves role-specific evidence, and uses the resulting context to construct a versioned PageSpec through a controlled generation workflow with deterministic and model-assisted routes. PageSpec explicitly represents page structure, states, interactions, acceptance conditions, and traceability links, and is deterministically rendered into an offline HTML/CSS/JavaScript prototype. Separate browser, structural, and requirement-level checks produce validation evidence without modifying the generated PageSpec or prototype, while a traceable result package and Inspector preserve evidence, generation records, failures, and fallback decisions for inspection and replay. In focused evaluations, Req2Web satisfied 16 of 18 sequential interaction checks, compared with 8 of 18 for direct HTML generation and 5 of 18 for one-shot structured generation. It also satisfied all 18 cross-artifact traceability checks, while an external rule engine reached the same coverage after equivalent cross-artifact rules were explicitly provided.