Add local web interface for resume screening

.github/workflows/build-windows-portable.yml

@@ -0,0 +1,165 @@
+name: Build Windows Packages
+
+on:
+  workflow_dispatch:
+    inputs:
+      bundle_local_model:
+        description: "Build a full offline installer with the Qwen GGUF model bundled"
+        required: false
+        type: boolean
+        default: false
+  push:
+    branches:
+      - resume-screening-cli
+    paths:
+      - ".github/workflows/build-windows-portable.yml"
+      - "packaging/**"
+      - "scripts/build_windows_app.ps1"
+      - "src/**"
+      - "tests/**"
+      - "web/**"
+      - "pyproject.toml"
+
+jobs:
+  build-windows-portable:
+    runs-on: windows-latest
+    timeout-minutes: 90
+    env:
+      DEFAULT_QWEN_MODEL_FILENAME: Qwen3.5-9B-Q4_K_M.gguf
+      DEFAULT_QWEN_MODEL_URLS: |
+        https://huggingface.co/unsloth/Qwen3.5-9B-GGUF/resolve/main/Qwen3.5-9B-Q4_K_M.gguf?download=true
+        https://huggingface.co/lmstudio-community/Qwen3.5-9B-GGUF/resolve/main/Qwen3.5-9B-Q4_K_M.gguf?download=true
+        https://modelscope.cn/models/diodel/Qwen3.5-9B-Q4_K_M-GGUF/resolve/master/Qwen3.5-9B-Q4_K_M.gguf
+      BUNDLE_LOCAL_MODEL: ${{ github.event_name == 'workflow_dispatch' && inputs.bundle_local_model && '1' || '0' }}
+      WINDOWS_ARTIFACT_NAME: ${{ github.event_name == 'workflow_dispatch' && inputs.bundle_local_model && 'xiaoa-resume-screening-windows-x64-offline' || 'xiaoa-resume-screening-windows-x64' }}
+
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Set up Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: "20"
+          cache: "npm"
+          cache-dependency-path: "web/package-lock.json"
+
+      - name: Install Inno Setup
+        shell: pwsh
+        run: choco install innosetup --no-progress --yes
+
+      - name: Build llama.cpp server runtime
+        shell: pwsh
+        run: |
+          git clone --depth 1 https://github.com/ggml-org/llama.cpp.git "$env:RUNNER_TEMP\llama.cpp"
+          cmake -S "$env:RUNNER_TEMP\llama.cpp" -B "$env:RUNNER_TEMP\llama-build" -DGGML_NATIVE=OFF -DGGML_OPENMP=OFF -DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_EXAMPLES=ON
+          cmake --build "$env:RUNNER_TEMP\llama-build" --config Release --target llama-server --parallel 2
+          New-Item -ItemType Directory -Force packaging\runtime\windows-x64 | Out-Null
+          Copy-Item "$env:RUNNER_TEMP\llama-build\bin\Release\llama-server.exe" packaging\runtime\windows-x64\
+          Copy-Item "$env:RUNNER_TEMP\llama-build\bin\Release\*.dll" packaging\runtime\windows-x64\ -ErrorAction SilentlyContinue
+
+      - name: Download bundled Qwen model
+        if: env.BUNDLE_LOCAL_MODEL == '1'
+        shell: pwsh
+        run: |
+          $targetDir = "packaging\models\qwen"
+          New-Item -ItemType Directory -Force $targetDir | Out-Null
+          $target = Join-Path $targetDir $env:DEFAULT_QWEN_MODEL_FILENAME
+          if (-not (Test-Path $target)) {
+            $urls = $env:DEFAULT_QWEN_MODEL_URLS -split "`r?`n" | ForEach-Object { $_.Trim() } | Where-Object { $_ }
+            $errors = @()
+            foreach ($url in $urls) {
+              Write-Host "Downloading bundled Qwen model from $url"
+              if (Test-Path $target) {
+                Remove-Item $target -Force
+              }
+              curl.exe -L --fail --retry 5 --retry-delay 10 --retry-all-errors --connect-timeout 30 --max-time 2400 --output $target $url
+              if ($LASTEXITCODE -ne 0) {
+                $errors += "$url failed with curl exit code $LASTEXITCODE"
+                if (Test-Path $target) {
+                  Remove-Item $target -Force
+                }
+                continue
+              }
+              $size = (Get-Item $target).Length
+              if ($size -ge 1000000000) {
+                break
+              }
+              $errors += "$url downloaded a too-small file: $size bytes"
+              Remove-Item $target -Force
+            }
+            if (-not (Test-Path $target)) {
+              throw "Failed to download bundled model. Attempts: $($errors -join '; ')"
+            }
+          }
+          $size = (Get-Item $target).Length
+          if ($size -lt 1000000000) {
+            throw "Downloaded model looks too small: $size bytes"
+          }
+          Write-Host "Bundled model ready: $target ($size bytes)"
+
+      - name: Run backend tests
+        run: python -m pip install ".[test,desktop]" && python -m pytest -q
+
+      - name: Build Windows portable and installer packages
+        shell: pwsh
+        run: |
+          $arguments = @{ Python = "python"; Installer = $true }
+          if ($env:BUNDLE_LOCAL_MODEL -eq "1") {
+            $arguments.BundleLocalModel = $true
+          }
+          .\scripts\build_windows_app.ps1 @arguments
+
+      - name: Verify Windows portable package
+        shell: pwsh
+        run: |
+          $runtime = "dist\小A简历筛选\_internal\packaging\runtime\windows-x64\llama-server.exe"
+          if (-not (Test-Path $runtime)) {
+            throw "Windows portable package is missing $runtime"
+          }
+          Add-Type -AssemblyName System.IO.Compression.FileSystem
+          $zipPath = "dist\小A简历筛选-windows-x64.zip"
+          $zip = [System.IO.Compression.ZipFile]::OpenRead($zipPath)
+          try {
+            $match = $zip.Entries | Where-Object { $_.FullName -like "*packaging/runtime/windows-x64/llama-server.exe" } | Select-Object -First 1
+            if (-not $match) {
+              throw "Windows portable zip is missing packaging/runtime/windows-x64/llama-server.exe"
+            }
+            if ($env:BUNDLE_LOCAL_MODEL -eq "1") {
+              $modelEntry = $zip.Entries | Where-Object { $_.FullName -like "*packaging/models/qwen/$env:DEFAULT_QWEN_MODEL_FILENAME" } | Select-Object -First 1
+              if (-not $modelEntry) {
+                throw "Windows portable zip is missing bundled local model"
+              }
+            }
+          } finally {
+            $zip.Dispose()
+          }
+          $setup = "dist\小A简历筛选-windows-x64-setup.exe"
+          if (-not (Test-Path $setup)) {
+            throw "Windows installer package is missing $setup"
+          }
+          if ($env:BUNDLE_LOCAL_MODEL -eq "1") {
+            $setupBins = Get-ChildItem -Path dist -Filter "小A简历筛选-windows-x64-setup*.bin"
+            if (-not $setupBins) {
+              throw "Windows offline installer is missing setup bin slices"
+            }
+            $model = "dist\小A简历筛选\_internal\packaging\models\qwen\$env:DEFAULT_QWEN_MODEL_FILENAME"
+            if (-not (Test-Path $model)) {
+              throw "Windows package is missing bundled local model $model"
+            }
+          }
+
+      - name: Upload Windows packages
+        uses: actions/upload-artifact@v4
+        with:
+          name: ${{ env.WINDOWS_ARTIFACT_NAME }}
+          path: |
+            dist/小A简历筛选-windows-x64.zip
+            dist/小A简历筛选-windows-x64-setup.exe
+            dist/小A简历筛选-windows-x64-setup*.bin
+          if-no-files-found: error

.gitignore

@@ -1,8 +1,16 @@
 .scratch/
 .worktrees/
+.superpowers/
 node_modules
+__pycache__/
+*.egg-info/
+.DS_Store
 data/
 logs/
 outputs/
+build/
+dist/
+packaging/models/
+*.spec.bak
 .env
 config.yaml

README.md

@@ -0,0 +1,81 @@
+# Resume Screening CLI
+
+## Setup
+
+```bash
+python3 -m venv .venv
+source .venv/bin/activate
+python3 -m pip install ".[test]"
+cp config.example.yaml config.yaml
+```
+
+Edit `config.yaml` and set `model.base_url`, `model.api_key`, and `model.model`.
+
+If the resume filename job name differs from the job workbook sheet name, add aliases:
+
+```yaml
+job_aliases:
+  后端开发工程师: "全栈"
+  后端开发实习岗: "全栈"
+```
+
+## Run
+
+```bash
+resume-screening run --config config.yaml
+```
+
+The tool appends rows to `/Users/mac/Downloads/小A科技（北京）组织招聘.xlsx`.
+It rereads `/Users/mac/Downloads/小A自动化岗位说明书_副本.xlsx` on every run.
+Historical rows are preserved.
+Duplicate resume content is still appended and marked red.
+The resume folder is scanned recursively. All visible regular files are processed; unsupported formats are appended as `待人工二筛` with an extraction note.
+Supported filename formats include `【岗位_地点 薪资】姓名 工龄.pdf` and `岗位_地点_薪资_姓名_工龄.pdf`.
+The job workbook can use the original one-sheet-per-job layout or a tabular layout where each row contains a `岗位名称`.
+
+## Web Interface
+
+Double-click `启动简历筛选网页版.command` from this project folder.
+
+On first launch, the script creates or reuses `.venv`, installs the tool, creates `config.yaml` from `config.example.yaml` if needed, builds the React page if needed, and opens `http://127.0.0.1:8765`.
+
+The web page has four steps:
+
+1. Save model settings.
+2. Confirm resume folder, job requirements workbook, result workbook paths, and optional job aliases.
+3. Run a lightweight precheck.
+4. Start screening and watch live progress, logs, and final statistics.
+
+## Local Desktop App
+
+Build a portable macOS app bundle:
+
+```bash
+source .venv/bin/activate
+bash scripts/build_macos_app.sh
+```
+
+The outputs are `dist/小A简历筛选.app` and `dist/小A简历筛选-mac.zip`.
+
+The app stores its config at `~/Library/Application Support/小A简历筛选/config.yaml`.
+When copied to a new Mac, copy `小A简历筛选-mac.zip`, unzip it, launch the app, and reselect the local resume folder, job workbook, result workbook, and model API settings.
+The app bundles the Python runtime and project dependencies. OCR still depends on a local `tesseract` installation if image or scanned-PDF recognition is needed.
+The app is built for the architecture of the Mac that runs the build script. The current generated package is for Apple Silicon; build again on Intel Mac if an Intel-only Mac needs to run it.
+
+Build Windows packages on Windows:
+
+```powershell
+.\scripts\build_windows_app.ps1 -Python python -Installer
+```
+
+The outputs are `dist\小A简历筛选-windows-x64.zip` and `dist\小A简历筛选-windows-x64-setup.exe`.
+The installer checks WebView2 and can download it after the user confirms the selected installer task.
+Qwen model weights are not bundled; the app checks the local model environment on first launch and downloads the GGUF model only after the user confirms in the app.
+
+## Manual-Review Mode
+
+By default, `model.fallback_to_local_when_unavailable: true` lets the web app switch from an unavailable custom OpenAI-compatible model to the bundled local Qwen model during precheck.
+
+Set `model.allow_without_model: true` to process files without model calls.
+Rows that require model judgment are classified as `待人工二筛`.
+When `base_url`, `api_key`, and `model` are all set, the tool uses the model first even if fallback mode is enabled.

config.example.yaml

@@ -0,0 +1,51 @@
+resume_dir: "/Users/mac/Downloads"
+job_book: "/Users/mac/Downloads/小A自动化岗位说明书_副本.xlsx"
+result_book: "/Users/mac/Downloads/小A科技（北京）组织招聘.xlsx"
+
+job_aliases: {}
+# 示例：
+# job_aliases:
+#   后端开发工程师: "全栈"
+#   后端开发实习岗: "全栈"
+
+job_profile_overrides: {}
+# 预检页确认岗位画像后会自动写入：
+# job_profile_overrides:
+#   全栈: "自定义岗位核心画像..."
+
+default_source_channel: ""
+default_interviewer: ""
+index_path: "data/processed_index.json"
+ocr_command: "tesseract"
+
+model:
+  provider: "openai-compatible"
+  base_url: ""
+  api_key: ""
+  model: ""
+  timeout_seconds: 60
+  temperature: 0.1
+  allow_without_model: false
+  fallback_to_local_when_unavailable: true
+  local:
+    runtime: "llama.cpp"
+    model_family: "qwen3.5"
+    model_display_name: "Qwen3.5 本地模型"
+    model_path: ""
+    manifest_path: "models/qwen/manifest.json"
+    host: "127.0.0.1"
+    port: 18080
+    context_size: 8192
+    threads: 0
+    gpu_layers: "auto"
+    auto_start: true
+    auto_download: false
+
+screening:
+  score_pass: 7
+  score_excellent: 9
+  categories:
+    recommend: "推荐初试"
+    consider: "可考虑"
+    manual: "待人工二筛"
+    reject: "未通过"