I’ve been working on VizuLLM, an open-source engine that turns structured LLM outputs (JSON) into deterministic, printable documents like timetables, invoices, guides, and diagrams.
LLMs are great at generating text, but in real applications we usually need documents, not paragraphs. VizuLLM introduces a strict contract between the model and the UI: • LLM generates schema-validated JSON • Schemas define a document type • VizuLLM renders a deterministic visual artifact (HTML / PDF)
Instead of asking an LLM to “write a PDF”, you ask it to generate data for a document schema, and the renderer handles the rest.
What makes it different: • Schema-driven (Zod) → predictable outputs • Renderer-first, not prompt-first • Print-optimized components • Open-source and community-extendable (anyone can add new document types)
Example use cases: • Generate schedules, invoices, cover letters • Turn LLM output into guides, reports, or diagrams • Integrate into internal tools or LLM workflows
Repo: https://github.com/GenR8ive/VizuLLM
I’d love feedback on: • Whether this abstraction makes sense • Missing document types / use cases
Happy to answer questions or discuss tradeoffs.