DocVision AI
Document Intelligence Operating System — turn any PDF/scan/DOCX into a structured, verifiable and versioned document.
Three layers
- Document Tools — inspect, extract text (with provenance), search, and operate on pages (rotate, merge, split, extract, reorder, delete). Real (pdf-lib/pdfjs).
- DocVision Intelligence — OCR consensus + confidence, review queue + heatmap, quality gates, round-trip, structure, semantic/visual diff, field extraction + provenance, redaction, forensics. Deterministic.
- Document Operating System — immutable documents and versions, patches, fields, annotations, reviews, quality reports, relations, audit. Persistent.
Integrity rules
- Never complete an illegible word/number → UNKNOWN.
- Uncertainty is measured: green / yellow / red / gray.
- Every value traces back to its source (page/area).
- The original is preserved; every edit = a patch → new version.
- Forensics flags anomalies without concluding forgery.
Real vs deferred (honest)
Real now: PDF operations and deterministic engines + persistent Document OS.
Later worker layer: real multi-engine raster OCR of scans, image restoration, .docx binary generation, real redaction burn-in, large-scale batch, RBAC/tenants.
Try it
MCP endpoint: /mcp (or /api/mcp). Call docvision_apercu, then docvision_inspect / docvision_extract_text with { "source": { "url": "https://…pdf" } }.