[4/12] feat: intelligence layer — diarization, LLM cleanup, runtime qualification, benchmarks

windows-native/Muesli.Windows/Muesli.Windows.csproj

@@ -20,6 +20,8 @@
     <Page Include="App.xaml" />
   </ItemGroup>
   <ItemGroup>
+    <PackageReference Include="LLamaSharp" Version="0.27.0" />
+    <PackageReference Include="LLamaSharp.Backend.Cpu" Version="0.27.0" />
     <PackageReference Include="NAudio" Version="2.2.1" />
     <PackageReference Include="org.k2fsa.sherpa.onnx" Version="1.13.4" />
     <PackageReference Include="QuestPDF" Version="2026.5.0" />

windows-native/Muesli.Windows/Services/BenchmarkCaptureRetentionPolicy.cs

@@ -0,0 +1,13 @@
+namespace Muesli.Windows.Services;
+
+public static class BenchmarkCaptureRetentionPolicy
+{
+    public const long MaximumBytes = 96L * 1024 * 1024;
+    public const int MaximumDurationMs = 10 * 60 * 1000;
+
+    public static bool ShouldRetain(long byteLength, int durationMs) =>
+        byteLength > 44 &&
+        byteLength <= MaximumBytes &&
+        durationMs > 0 &&
+        durationMs <= MaximumDurationMs;
+}

windows-native/Muesli.Windows/Services/FillerWordFilter.cs

@@ -0,0 +1,68 @@
+using System.Text.RegularExpressions;
+
+namespace Muesli.Windows.Services;
+
+public static partial class FillerWordFilter
+{
+    private static readonly string[] PhraseFillers =
+    [
+        "you know",
+        "i mean",
+        "sort of",
+        "kind of"
+    ];
+
+    private static readonly string[] WordFillers =
+    [
+        "uh", "um", "uhh", "umm", "er", "err", "ah", "ahh", "hmm", "hm", "mm", "mmm"
+    ];
+
+    public static string Apply(string text)
+    {
+        if (string.IsNullOrWhiteSpace(text))
+        {
+            return text;
+        }
+
+        var filtered = text;
+        filtered = Regex.Replace(
+            filtered,
+            @"(?<![\p{L}\p{N}])like\s*,(?![\p{L}\p{N}])",
+            "",
+            RegexOptions.IgnoreCase | RegexOptions.CultureInvariant);
+        foreach (var phrase in PhraseFillers)
+        {
+            filtered = Regex.Replace(
+                filtered,
+                $@"(?<![\p{{L}}\p{{N}}]){Regex.Escape(phrase)}(?:\s*,)?(?![\p{{L}}\p{{N}}])",
+                "",
+                RegexOptions.IgnoreCase | RegexOptions.CultureInvariant);
+        }
+
+        foreach (var word in WordFillers)
+        {
+            filtered = Regex.Replace(
+                filtered,
+                $@"(?<![\p{{L}}\p{{N}}]){Regex.Escape(word)}(?:\s*,)?(?![\p{{L}}\p{{N}}])",
+                "",
+                RegexOptions.IgnoreCase | RegexOptions.CultureInvariant);
+        }
+
+        filtered = WhitespaceBeforePunctuation().Replace(filtered, "$1");
+        filtered = RepeatedWhitespace().Replace(filtered, " ").Trim();
+        if (filtered.Length == 0)
+        {
+            return filtered;
+        }
+
+        return char.IsLower(filtered[0])
+            ? char.ToUpperInvariant(filtered[0]) + filtered[1..]
+            : filtered;
+    }
+
+    [GeneratedRegex(@"\s+([,.;:!?])", RegexOptions.CultureInvariant)]
+    private static partial Regex WhitespaceBeforePunctuation();
+
+    [GeneratedRegex(@"\s+", RegexOptions.CultureInvariant)]
+    private static partial Regex RepeatedWhitespace();
+}