AI Labs GitHub
Table of contents

POE Verification Analyst

App3 min read

On-device proof-of-execution verification: for each case it cross-checks the invoice against the purchase order and the delivery / execution evidence, flags exposure, and detects evidence reused across claims.

How it works

flowchart LR
  DOCS[PO, invoice, delivery note PDFs] -->|pdfText / ocr| EX[extractDocumentFields<br/>model → JSON, validated]
  EX --> MX[buildMatrix<br/>quantities, prices, dates, parties]
  MX --> EXP[computeExposure + computeRiskIndicators]
  PH[photos] -->|describeImage| AP[analyzePhotos<br/>quantity estimate]
  EX -->|indexPoe op| IDX[(AppContentIndexer)]
  IDX -->|queryPoe op| RU[searchReuse + client-side confirm filter]
  EXP & AP & RU --> CP[case copilot]
  • The verification matrix is deterministic (buildMatrix), the model only extracts fields (buildExtractionPrompt, validateExtractedFields, coerceNumber) and explains findings.
  • Reuse detection needs a retry/delay because the index is asynchronous, and a client-side confirmation step because the index matches fuzzily (reuseMatches).
  • Photo quantity estimation is a model estimate (analyzePhotos), shown as such.
  • Supplier view: supplierStats, renderSuppliers.

Source files

File What it contains
MainWindow.xaml.cs Bridge ops ensureReady, generate, pdfText, ocr, describeImage, indexPoe, queryPoe; DecodeImageBufferAsync, GetPoeIndex.
www/index.html UI: Dashboard, New Case, Cases, Suppliers, Local AI.
www/app.js Logic (≈39 KB): extraction prompts and validation, matrix, exposure/risk, reuse search, supplier stats, copilot.
manual-test-documents/ Six synthetic scenarios + generate.py + README: clean, quantity over-billing, price inflation, execution shortfall, back-dated evidence, photo discrepancy.

Step-by-step deployment

Everything below is copy-paste PowerShell on a Windows 11 PC. Replace nothing: the repository owner is medyabi.

1. Check the PC and install the tools (once)

  • Windows 11 build 22621 or later, x64 (winver). For the AI features you need a Copilot+ PC (NPU); otherwise the app starts but reports the model as unavailable.
PowerShell
winget install --id Microsoft.DotNet.SDK.8
winget install --id Git.Git
winget install --id GitHub.cli

2. Sign in to GitHub and clone (the repository is private, so the account must have access)

PowerShell
gh auth login
git clone https://github.com/medyabi/windows-local-ai-apps.git
cd windows-local-ai-apps

3. Build the signed package (about 30 s to 2 min; the first build also downloads NuGet packages and needs internet)

PowerShell
.\scripts\build-app.ps1 -App POEVerificationApp

It creates a self-signed certificate if needed and writes two files: dist\POEVerificationApp\POEVerificationApp_1.0.0.0_x64.msix and dist\POEVerificationApp\POEVerificationApp.cer. A warning about mspdbcmf.exe is harmless.

4. Trust the certificate (PowerShell as administrator, once per PC)

PowerShell
Import-Certificate -FilePath .\dist\POEVerificationApp\POEVerificationApp.cer -CertStoreLocation Cert:\LocalMachine\TrustedPeople

5. Install

PowerShell
Add-AppxPackage -Path .\dist\POEVerificationApp\POEVerificationApp_1.0.0.0_x64.msix

6. Launch it from the Start menu (POE Verification Analyst) or:

PowerShell
$p = Get-AppxPackage *POEVerification*
Start-Process "shell:AppsFolder\$($p.PackageFamilyName)!App"

7. First launch

  • The status pill shows the on-device model state. The first time, Windows may need to prepare the model (a few minutes, may need Windows Update).
  • Use the scenario folders in manual-test-documents/ (supplier names are listed in its README). Extraction takes about 5-15 s per document.

8. If something fails

PowerShell
$p = Get-AppxPackage *POEVerification*
notepad "$env:LOCALAPPDATA\Packages\$($p.PackageFamilyName)\LocalState\bridge.log"

Common errors and fixes are in docs/DEPLOYMENT.md.

9. Uninstall (also removes the app's data)

PowerShell
Get-AppxPackage *POEVerification* | Remove-AppxPackage

Try it

  1. New Case → type the supplier name from the table in manual-test-documents/README.md (e.g. Atlas Logistics) → drop that folder's PO.pdf, Invoice.pdf, POE.pdf → Extract & Verify (about 5–15 s per document).
  2. Open the case: scenario 2 should show a quantity mismatch (PO 300 / invoice 340 / POE 300) and an exposure of about €8,000; scenario 1 should be clean.
  3. In an open case use Search for reused evidence, ask the Investigation Copilot, and for scenario 6 click Analyze photos with on-device AI.

Limits

Synthetic documents only; extraction quality on real paperwork is untested. The page loads a Google Fonts stylesheet (cosmetic).

View the source on GitHub