HR Recruiter Agent
An on-device recruiting copilot: create a role from a job description, drop résumés, and get an evidence-backed fit assessment per candidate.
How it works
flowchart LR JD[job description<br/>PDF / paste / URL] --> RA[analyzeRole<br/>1 model call] RES[résumé PDFs] -->|pdfText| PR[processResume] PR --> P[1. profile] P --> F[2. fit vs role + evidence quotes] F --> I[3. insights / interview questions] RA --> F PR -->|indexResume op| IDX[(AppContentIndexer)] IDX --> DUP[checkDuplicates] F & I --> CP[copilot chat per role / candidate]
- The work is deliberately split into several small model calls (
buildProfilePrompt,buildFitPrompt,buildInsightsPrompt), each validated (validateProfile,validateFit,validateInsights,extractJsonLoose). One big combined call took about 8 minutes and truncated its JSON. - Evidence: the fit assessment quotes résumé text;
renderCandidateProfileshows it next to each requirement. - Duplicate detection uses the local semantic index (
indexResume/queryResume). - A Responsible AI page states the limits.
Source files
| File | What it contains |
|---|---|
MainWindow.xaml.cs |
Bridge ops ensureReady, generate, pdfText, indexResume, queryResume, fetchUrl. HandleFetchUrlAsync downloads a web page (HTML) with HttpClient because the browser's CORS rules would block it in the page. |
www/index.html |
UI: Dashboard, New Role, Roles, Responsible AI. |
www/app.js |
Logic (≈48 KB): prompts, validation, role/candidate rendering, duplicate check, copilot context builders. |
sample-test-documents/ |
One job description and two synthetic résumés, with a README of expected results. |
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 HRRecruiterAppIt creates a self-signed certificate if needed and writes two files: dist\HRRecruiterApp\HRRecruiterApp_1.0.0.0_x64.msix and dist\HRRecruiterApp\HRRecruiterApp.cer. A warning about mspdbcmf.exe is harmless.
4. Trust the certificate (PowerShell as administrator, once per PC)
Import-Certificate -FilePath .\dist\HRRecruiterApp\HRRecruiterApp.cer -CertStoreLocation Cert:\LocalMachine\TrustedPeople5. Install
Add-AppxPackage -Path .\dist\HRRecruiterApp\HRRecruiterApp_1.0.0.0_x64.msix6. Launch it from the Start menu (HR Recruiter Agent) or:
$p = Get-AppxPackage *HRRecruiterAgent*
Start-Process "shell:AppsFolder\$($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 files in
sample-test-documents/. - Only the "fetch job description from URL" feature needs internet.
8. If something fails
$p = Get-AppxPackage *HRRecruiterAgent*
notepad "$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 *HRRecruiterAgent* | Remove-AppxPackageTry it
- New Role → drop
sample-test-documents/JobDescription-SeniorCloudEngineer.pdf→ Analyze. - Open the role → drop the two sample résumés; compare the strong and weak fit.
- Ask the copilot why a candidate was ranked as they were.
Limits
- One feature uses the network: "fetch job description from URL" downloads that page. Everything else is local. The page also loads a Google Fonts stylesheet.
- Do not use the host's DevTools relay with this app: it crashed it regardless of load.
- A model must never be the only judge of a candidate: demo only.