Editorial workstation
The adaptive interview extracts evidence; the hybrid gateway (LLM + heuristics) blocks invented lived experience. Only confirmed content moves to draft and Markdown or SlideMark export.
Entrevista extrai evidência; briefing e storyboard preparam o material; o gateway veta fabricado. Só o confirmado segue para rascunho e export.
Generating a “professional” post with an LLM from a slogan invents biography. The flow must interview, identify gaps, prepare the briefing, draft, and export — and refuse what was not confirmed.
Python core with boundaries between interview, generation, authorship preservation, segmentation, and persistence. Model calls in an isolated workspace, provider allowlist, and prompts as versioned contracts. Hybrid evaluation: LLM plus deterministic heuristics. Missing experience never becomes false lived experience.
Adaptive interviews, authorial briefing, storyboard, veto on fabricated content, SQLite prompt registry with rollback, textual interface, and React/Vite frontend to review phases. Markdown or SlideMark JSON export after evaluation.
Base with tests for interview, LLM isolation, registry, persistence, and SlideMark conversion.
It is not a generic thought-leadership generator. Without real repertoire, the system refuses; it does not complete the biography.