Local Fraud Investigator
An on-device fraud / AML investigation assistant. Load transactions and customer profiles, see risk scores and a transaction graph, open a case, and ask questions against your policy documents. The language model and the policy index run on the PC.
How it works
flowchart LR CSV[transactions + profiles CSV] --> A[analyzeAll<br/>z-scores, new receivers/countries, gaps] A --> D[Dashboard + risk list] D --> CU[openCustomer → runCustomerAssessment] POL[policy PDFs] -->|pdfText| CH[chunkText] CH -->|indexPolicy op| IDX[(AppContentIndexer)] CU --> Q[retrievePolicyContext] IDX -->|queryPolicy op| Q Q --> M[generate op Phi Silica] M --> R[assessment, case report, copilot answers]
- Deterministic analysis in
app.js: statistics (avg,stddev,z), first-time receivers/countries (knownReceiversBefore,knownCountriesBefore), time gaps, categories, a relationship graph (renderGraph). - Policy RAG: policy documents are chunked and added to a local semantic index; the retrieved passages are put in the prompt so the answer cites the policy.
- Natural-language query (
runNLQuery→executeStructuredQuery): the model turns a question into a structured filter that the code executes on the data. - Cases:
createCase,generateReport.
Source files
| File | What it contains |
|---|---|
MainWindow.xaml.cs |
Bridge ops ensureReady, generate, pdfText, ocr, indexPolicy, queryPolicy. GetPolicyIndex creates the AppContentIndexer; index calls run in Task.Run. |
www/index.html |
UI shell with three views: Dashboard, Cases, Policy Library. |
www/app.js |
All logic (≈50 KB): CSV parsing, analysis, graph, cases, policy indexing/retrieval, copilot chat. |
www/sample-transactions.csv, www/customer-profiles.csv, www/policies/*.pdf |
Synthetic data loaded by loadDemoDataset / loadDemoPolicies (AML, KYC, transaction-monitoring and investigation-handbook policies). |
sample-data/ |
The same kind of data as standalone files. |
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.
winget install --id Microsoft.DotNet.SDK.8
winget install --id Git.Git
winget install --id GitHub.cli2. Sign in to GitHub and clone (the repository is private, so the account must have access)
gh auth login
git clone https://github.com/medyabi/windows-local-ai-apps.git
cd windows-local-ai-apps3. Build the signed package (about 30 s to 2 min; the first build also downloads NuGet packages and needs internet)
.\scripts\build-app.ps1 -App FraudInvestigatorAppIt creates a self-signed certificate if needed and writes two files: dist\FraudInvestigatorApp\FraudInvestigatorApp_1.0.0.0_x64.msix and dist\FraudInvestigatorApp\FraudInvestigatorApp.cer. A warning about mspdbcmf.exe is harmless.
4. Trust the certificate (PowerShell as administrator, once per PC)
Import-Certificate -FilePath .\dist\FraudInvestigatorApp\FraudInvestigatorApp.cer -CertStoreLocation Cert:\LocalMachine\TrustedPeople5. Install
Add-AppxPackage -Path .\dist\FraudInvestigatorApp\FraudInvestigatorApp_1.0.0.0_x64.msix6. Launch it from the Start menu (Local Fraud Investigator) or:
$p = Get-AppxPackage *FraudInvestigator*
Start-Process "shell:AppsFolder\$($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).
- Dashboard: load the demo dataset. Policy Library: load the demo policies, then wait a few seconds for the asynchronous index before asking policy questions.
8. If something fails
$p = Get-AppxPackage *FraudInvestigator*
notepad "$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)
Get-AppxPackage *FraudInvestigator* | Remove-AppxPackageTry it
- Dashboard → load the demo dataset; open a high-risk customer and run the assessment.
- Policy Library → load the demo policies (wait a few seconds: the index is asynchronous), then ask the copilot a policy question.
- Create a case and generate its report.
Limits
The local index is asynchronous and fuzzy; calling it directly from the UI thread deadlocks (see docs/WINDOWS-AI-LESSONS.md). Synthetic data and demo policies only: not for real investigations. The page loads a Google Fonts stylesheet (cosmetic); transaction data is not sent anywhere.