User enters raw domain names via text input or CSV file upload. The system cleans and deduplicates the list for batch processing.

AI decomposes domain keywords, explores cross-language meanings, acronyms, and mashups. Generates 7-10 scored niche ideas with monetization models, audience profiles, and valuation bands.

User selects a niche, template type, layout, and density. A 4-step discovery modal captures brand personality, audience, values, and visual preferences. When skipped, _infer_discovery_from_niche() auto-generates rich context from niche data.

The Blueprint Engine assembles a site specification from 16 section types across 6 categories. Users choose a depth preset which controls how many sections are enabled, each with field specs and AI instructions.

The build engine transforms the blueprint + niche data + discovery context into a complete business package. GPT generates 3 brand identity options and full site copy for every enabled section in a single large JSON response.

Raw AI JSON output passes through a 3-layer validation and auto-repair pipeline. Handles field aliases, nested object unwrapping, list coercion, missing field repair, and type normalization to guarantee consistent downstream data.

Generates hero images (1536×1024) and section-specific images using DALL-E. Images are generated sequentially to respect API rate limits. Prompts are derived from brand data, niche context, and visual style preferences.

All validated data, images, and sales materials are assembled into the final output. Users get a live preview site, a visual editor for refinements, a standalone sales page, and a ZIP export containing all assets.