Mindcraft Book audit
Mindcraft / Capabilities / Document AI
Module 05 · Capability

Documents that
read themselves.

Contracts, invoices, applications, PDFs — parsed, summarised, validated, routed. Operations stops being a hand-drag through paperwork. Lawyers read one page, not forty.

What this fixes

Three reasons
paperwork wins.

If your team has ever pasted invoice fields into a spreadsheet at midnight, you're who we built this for.

Problem 01

Paperwork is the work.

Operations teams spend 40% of their time reading documents that all look mostly the same — invoices, contracts, applications, intake forms — and pasting fields into other tools.

Problem 02

Edge cases derail everything.

The agent who handles 200 invoices a day misses the one with a fraudulent total. The intake reviewer rubber-stamps the application that should have flagged. Volume creates blindness.

Problem 03

Searching is hell.

Need to know which contracts have the auto-renewal clause? Hope you saved them somewhere searchable. Hope you remember the right keyword. Hope you have an hour.

Sample analysis

What it
actually does.

An illustrative services agreement, modeled on common SaaS contract patterns. Below: the document on the left with the AI's clause highlights. The right shows the structured fields it extracted, with confidence scores and warnings on anything non-standard. Numbers and names are fictional.

Auto-analysed 8 fields extracted 3 flags raised Processed in 2.4s · uploaded 14:42 EST
Draft · v3

Master Service Agreement.

HEXAGRID INDUSTRIES · BRIGHT STUDIO INC. · 14 OCT 2025
1.Scope of services.

Provider shall deliver the services described in Schedule A, beginning on the Effective Date and continuing for the Initial Term as defined herein.

2.Payment terms.

Customer agrees to pay net 30 days from invoice receipt, in USD, billed monthly on the first business day. Late payments shall accrue interest at 1.5% per month.

3.Term & renewal.

Initial Term is twenty-four (24) months from Effective Date. Either party may terminate by providing thirty (30) days written notice prior to renewal.

5.Limitation of liability.

Provider's aggregate liability under this Agreement shall not exceed six (6) months of fees paid in the twelve (12) months preceding the claim. The parties acknowledge this cap as material consideration.

8.Governing law.

This Agreement shall be construed under the laws of the State of Delaware, United States, without regard to conflict-of-laws principles.

Extracted · structured
Confidence · 96%
Counter-party
Hexagrid Industries B2B SaaS · Toronto, CA · Series B
99%
0.2s
Payment terms
Net 30 · USD Monthly invoice · 1st business day · 1.5% late fee
99%
0.3s
Initial term
24 months Effective Oct 14, 2025 → Oct 13, 2027
98%
0.3s
Renewal notice
30 days Standard · matches template
97%
0.2s
Liability cap
6 months fees Below standard 12-month cap · review recommended
95%
0.4s
Jurisdiction
Delaware, USA No conflict-of-laws clause · standard
99%
0.2s
Auto-renewal
Yes Defaults to 12-month renewal · written notice req.
96%
0.3s
Total contract value
$180,000 $7,500/mo × 24 months · per Schedule A
92%
0.6s
AI plain-English summary

This is a standard 24-month services agreement with Hexagrid for $180,000 total contract value, billed monthly net-30. Renewal terms and jurisdiction are template-standard. One material concern: the liability cap is set at six months of fees rather than the twelve-month default — recommend pushing back during redlines, or accepting with a documented exception.

3 flags raised · vs your template
5 fields match standard
Routed to: legal queue · Sarah K.
Linked: SOW-2025-1014 · Hexagrid CRM
What ships

What's in the box.

Six things you get with every Document AI engagement. Real parsing, real validation, real summaries.

01

Parse anything you upload.

PDFs, scanned images, Word docs, emails-with-attachments, web forms. Tables, signatures, stamps, handwritten notes. The good and the messy. Twelve languages out of the box; more on request.

02

Extract what you actually need.

Specific fields — invoice total, vendor name, due date, payment terms, contract expiry, signatory, jurisdiction. Defined once, extracted forever. Confidence scores attached so you know what's trusted.

03

Validate against rules.

Total doesn't match line items? Flagged. Date inconsistent with PO? Flagged. Counter-party not on approved-vendor list? Flagged. Edge cases caught, not buried in someone's queue.

04

Summarise for humans.

Every contract gets a one-paragraph plain-English summary, and a one-page diff against your standard template. Lawyers read one page, not forty.

05

Searchable knowledge.

Every document indexed with semantic search. "Show me all contracts with auto-renewal under 90 days." Returned in one second, with sources. The archive becomes useful.

06

Route to the right place.

Invoice goes to AP. Contract goes to legal queue (or auto-files if standard). Application gets assigned by region. The document doesn't sit in someone's inbox waiting.

How it works

Four steps,
end to end.

Step 1

Doc arrives.

Email, upload, scanner, watched folder, or API. Whatever your team uses today, we plug into.

Step 2

Read + extracted.

Fields, structure, signatures, line items — all detected. Confidence scores attached.

Step 3

Validated + summarised.

Rules checked, plain-English digest written, diff against template generated.

Step 4

Routed + searchable.

To the right queue. Indexed forever. Findable in one second by anyone with permission.

Mission control

Today's intake,
in one screen.

Every document that came in, where it went, what was flagged, and what's still pending. The state of the paperwork in one screen.

MINDCRAFT · DOC INTAKE · TODAY Live · auto-refresh
Processed today
187
↑ 14 vs yesterday
Auto-validated
174
93% pass rate
Need review
13
3 high-priority
Avg time
2.4s
98.4% accuracy
MSA · Hexagrid Industries · v3 final
PDF · 12 pages · 487KB · Sarah K.
Contract
⚠ Review · 3 flags
2m ago
INV-2025-44801 · Ledger.io
PDF · 1 page · 84KB · vendor@
Invoice
✓ Filed · AP
14m ago
Vendor onboarding · Frostline LLC
PDF · 8 pages · 1.2MB · ops@
Application
Processing · 73%
just now
Receipt · AWS October billing
PDF · 4 pages · 220KB · finance@
Receipt
✓ Filed · AWS
28m ago
DPA · Lumen Bank · annotated
PDF · 18 pages · 2.4MB · legal@
DPA
⚠ Review · 2 flags
42m ago
Insurance certificate · Vellis renewal
PDF · 2 pages · 96KB · admin@
Form
✓ Filed · vendor
1h ago
INV-2025-1018-001 · Frostline
JPG (scan) · 1 page · 1.1MB
Invoice
✗ Total mismatch
1h ago
Built on

A stack you can audit, port, and own.

Documents stay in your environment. Claude with no-train guarantee. Optional VPC deployment for regulated industries. Standard NDA on engagement; DPA on request.

Claude (vision + extraction) pgvector / Pinecone Notion / Confluence n8n orchestration DocuSign / PandaDoc S3 / Drive / Dropbox
8 → 1

Tools collapsed at one pan-European retailer. A custom internal console replaced eight half-used SaaS subscriptions for invoice management, contract review, and document filing. Lower cost, faster onboarding, a UI built for the team that uses it daily.

Retail · 2025
Frequently asked

Questions we get a lot.

How accurate is the extraction?
On structured documents (invoices, intake forms): 97–99% field accuracy after one round of fine-tuning. On unstructured (contracts, emails): 92–95% for flagged fields. We always show confidence scores and route low-confidence to human review.
Can it handle our weird internal forms?
Yes. The whole point. We train on your forms, your layouts, your vocabulary. Most clients have a few documents that look "unprocessable" — we usually handle them. The audit covers a sample so you can see for yourself before signing.
What about confidentiality?
Documents stay in your environment. We use Claude with no-train guarantee, and can deploy to your VPC if needed. Standard NDA on engagement, plus optional DPA for European clients. SOC2-friendly architecture; we can map to your compliance requirements.
Is this going to put our ops team out of work?
No. It removes the boring 70% — typing things into systems. Your ops team moves up to handling exceptions, improving processes, and dealing with the hard cases that actually need a brain. Most teams report higher job satisfaction post-rollout.
Can it handle scans, photos, handwriting?
Yes — Claude's vision model handles all three. Quality depends on input quality (a photo taken with a thumb on the camera will struggle), but typical phone-scanned receipts and invoices are no problem. Handwritten signatures and notes are detected and extracted with confidence flags.
How long does setup take?
For a single document type (e.g., invoices): 1–2 weeks. For full intake across multiple types: 3–6 weeks. Week 1 is always discovery + sample analysis so you see the quality before committing further.
What's it cost?
Document AI is included in the Engine tier ($3,500/mo) for one document type. Additional types are $400/mo each. Pass-through Claude API costs typically $50–300/mo depending on volume and document complexity.
Often paired with

Related capabilities.

Send us
a sample document.

The free audit includes a live extraction on three of your real documents — anonymised if you want. You see the deliverable before you commit. We've yet to find a document we couldn't handle.

Book the audit