Building interm.ai – A Real-Time Helper with Electron, Swift & Bolt

---

⚡ TL;DR
interm.ai is my cross-platform AI assistant that listens to calls and whispers suggestions in real time—think Cluely, but laser-focused on interviews & sales demos.

• Electron desktop app for macOS / Windows

• Swift CLI on macOS for lossless system-audio + screenshots

• Google Speech-to-Text (Deepgram soon) → GPT-4o for sub-500 ms suggestions

• Landing page on Bolt.new; demo videos generated by Claude Code

---

🛠️ What We Built

---

🌐 Demos

Architecture Overview 🛠️

Swift CLI: Capturing macOS Audio & Screenshots 🎙️🖼️

// AudioScreenshot.swift
import AVFoundation
import CoreImage
import ScreenCaptureKit

@main
struct AudioScreenshot {
  static func main() async throws {
    // 1️⃣ Capture raw system audio
    let session = try SCStream.shared(systemAudio: true, microphone: false)

    // 2️⃣ Poll active window every 5 s
    while true {
      let image = try session.captureCurrentFrame()
      save(image)                       // → ~/Library/Caches/interm/frame.png
      try await Task.sleep(for: .seconds(5))
    }
  }
}

Roadmap 🗺️

1. On-device fallback with a tiny LLM for offline flights.

2. CRM webhook – push call highlights straight into HubSpot.

3. Switch speech backend to Deepgram once their low-latency best-word model exits beta.

4. Open-source the Swift CLI (after tidying up the build script).

---

Try it Out & Give Feedback 🙌

Try Interview Terminator