cashcrown // document analysis
Doc Intelligence
doc-intelliveA pipeline that turns messy documents — PDFs, scans, photos, DOCX — into structured, verifiable data: every field traceable to its source span, with a confidence threshold.
self-host
deploy
zero-retention
data
4 tryby
input sources
per-pole
confidence
// 01
Problem
Document data is either retyped by hand or pulled by an OCR that never tells you when it was wrong. A field with no source and no confidence lands in your system unchecked, and a single misread scan costs hours of corrections — or a silent overwrite of a value that mattered.
// 02
Capabilities
- OCR with a confidence score per field and per character
- document-type classification and field extraction
- every field traceable to its source span
- low-confidence fields routed to human review
- critical fields never auto-overwrite
// 05
Integrations
routerBGE-M3QdrantOllaman8nFastAPI
// 06
Architecture
Select a node to see its description and data flow.
Architecture — text description
- Ingest / data
- Data sources — Content, documents and events feed the index. (Flows to: BGE-M3)
- BGE-M3 — Local embeddings (1024-dim) — PII stays on-prem. (Flows to: OpenClaw router)
- AI core
- OpenClaw router — The only entry to the models — masks PII, picks the model, throttles. (Flows to: Doc Intelligence)
- Doc Intelligence — A pipeline that turns messy documents — PDFs, scans, photos, DOCX — into structured, verifiable data: every field traceable to its source span, with a confidence threshold. (Flows to: API)
- Serve / integrations
- API — REST surface — cited answers with access control. (Flows to: router, BGE-M3, Qdrant, Ollama)
- router — Integration: router.
- BGE-M3 — Integration: BGE-M3.
- Qdrant — Integration: Qdrant.
- Ollama — Integration: Ollama.
// 07
Live demo
// 08
FAQ
How much does it cost, how do you measure ROI, and how accurate is it?
We work in ranges that depend on scope — we start with a fixed-cost pilot on your own sample of documents, so you see the real accuracy before a larger investment. We quote accuracy as ranges tied to source quality: clean digital PDFs usually 95–99% of fields, good scans 90–97%, photos and poor scans 80–92% — which is exactly why low-confidence fields still go to review. When the pipeline takes back a dozen to several dozen hours of retyping a month, it usually pays off in 2–4 months. Calculate the return in the ROI calculator.
Does it integrate with our systems and is it GDPR-compliant?
Yes. We return results through an API and webhooks (n8n), plug into your document flow and CRM, and irreversible actions — including overwriting a critical field — require human confirmation (a human-gate). The pipeline is self-hosted and runs zero-retention: documents with PII are processed locally, nothing is kept after processing, and GDPR and AI Act compliance are designed in from the start.
Where do we start, and what documents can it handle?
We start with one document type and a few dozen of your real examples — enough to measure accuracy and set the confidence thresholds. We handle four input modes: digital PDFs, scans, photos, and DOCX (invoices, contracts, forms, reports). After the pilot we expand to more types and fields. We do not promise 100% automation — the goal is confident extraction where the model is sure, and human review where it is not.
// →Related