Interview AI

A macOS overlay app that listens to interview questions in real time and instantly displays your prepared answers.

一款 macOS 浮窗应用，实时监听面试问题并即时展示你预先准备的答案。

What is InterviewAI / 项目介绍

InterviewAI captures audio during interviews (system audio from Zoom/Teams/Meet or microphone input), transcribes the interviewer's questions using Apple's Speech framework, and matches them against your pre-written prompt library. The best-matching answer is displayed in a transparent overlay window — always on top, always ready.

InterviewAI 在面试过程中捕获音频（来自 Zoom/Teams/Meet 的系统音频或麦克风输入），利用 Apple Speech 框架将面试官的问题转录为文本，并与你预先编写的提示库进行匹配。匹配度最高的答案会显示在透明浮窗中——始终置顶，随时可用。

Features / 功能特性

• Real-time speech recognition — Powered by SFSpeechRecognizer, supports 30+ languages
• System audio capture — Captures app audio output via ScreenCaptureKit, no need for virtual audio devices
• Local matching — All processing happens on-device, no network requests, no data leaves your Mac
• Customizable overlay — Adjust opacity, font size, colors, and always-on-top behavior
• Prompt management — Create, edit, reorder, and search your Q&A library; import via JSON
• Lock answer — Pin the current answer to prevent it from changing while you're reading

---

• 实时语音识别 — 基于 SFSpeechRecognizer，支持 30+ 种语言
• 系统音频捕获 — 通过 ScreenCaptureKit 捕获应用音频输出，无需虚拟音频设备
• 本地匹配 — 所有处理在设备端完成，无网络请求，数据不离开你的 Mac
• 自定义浮窗 — 调整透明度、字体大小、颜色和置顶行为
• 提示管理 — 创建、编辑、排序、搜索问答库；支持 JSON 导入
• 锁定答案 — 固定当前答案，阅读时不会被新匹配替换

How It Works / 工作流程

flowchart LR
    A["Audio Input\n(System/Mic)"] --> B["Speech\nRecognition"] --> C["N-gram/IDF\nMatching"] --> D["Overlay\nDisplay"]

Loading

1. Capture — Audio is captured from system output (ScreenCaptureKit) or microphone
2. Transcribe — SFSpeechRecognizer converts speech to text in real time
3. Match & Display — The transcribed text is matched against your prompt library and the best answer is shown

Matching Algorithm / 匹配算法

InterviewAI uses a character-level N-gram matching algorithm with IDF weighting, designed to work across languages without requiring tokenization or word boundaries.

InterviewAI 使用字符级 N-gram 匹配算法，结合 IDF 加权，无需分词即可跨语言工作。

Indexing Phase / 索引阶段

When an interview starts, all prompts are pre-indexed:

面试开始时，所有提示会被预索引：

• Text is normalized (lowercased, punctuation/symbols removed)
• N-grams (2, 3, 4-character sliding windows) are extracted from titles and content
• Character sets are built for fast single-character lookup
• IDF scores are computed per character across all prompt titles:

IDF(char) = log(1 + totalDocs / docFreq(char))

Characters that appear in fewer prompts get higher weight — rare characters are more informative for matching.

出现在更少提示中的字符获得更高权重——稀有字符对匹配更有价值。

Scoring Phase / 评分阶段

Each recognized question is scored against every prompt using three layers:

每个识别到的问题通过三层评分与所有提示进行对比：

Layer	Description	Weight
Substring match	Full title found inside the question	titleLength × 20
N-gram overlap	Shared 2/3/4-grams between question and prompt, IDF-weighted	Title: × 10, Content: × 2
Character overlap	Shared individual characters, IDF-weighted	Title: × 3, Content: × 0.5

Title matches are weighted 5× higher than content matches, since interview questions are most likely to overlap with prompt titles.

标题匹配的权重是内容匹配的 5 倍，因为面试问题最可能与提示标题重合。

Confidence Threshold / 置信度阈值

A confidence score is calculated as:

confidence = score / maxPossibleScore

Results below 15% confidence are discarded. This prevents weak or irrelevant matches from appearing.

低于 15% 置信度 的结果会被丢弃，防止弱匹配或无关匹配出现。

Requirements / 系统要求

• macOS 14.0 (Sonoma) or later
• Xcode 16.0 or later
• Permissions required:
  • Microphone — for mic-based recognition
  • Speech Recognition — for on-device transcription
  • Screen Recording — for system audio capture (ScreenCaptureKit)

Build & Run / 编译运行

git clone https://github.com/RyuApps/interviewai.git
cd interviewai
open interviewai.xcodeproj

Select the interviewai scheme and run (⌘R). No dependencies, no package managers — pure Swift and system frameworks.

无第三方依赖，纯 Swift + 系统框架，克隆后直接编译运行。

Project Structure / 项目结构

interviewai/
├── Models/                 # Data models (PromptItem, AppSettings, MatchResult)
├── Views/                  # SwiftUI views (MainView, SettingsView, PromptEditor)
├── ViewModels/             # Observable view models (MVVM)
├── Services/               # Core services (Speech, SystemAudio, AI matching, Storage)
├── Managers/               # Session orchestration (InterviewSessionManager)
├── Utilities/              # Constants, logging
└── Resources/              # Sample prompts (JSON)

License / 许可证

MIT License — free to use, modify, and distribute.

MIT 许可证——可自由使用、修改和分发。

Download / 下载

This is the open-source edition of InterviewAI. The App Store version is also available.

这是 InterviewAI 的开源版本，同时也提供 App Store 版本。