AI & Intelligent Systems
BetaIn-house R&D · 2026
BrainCue — Real-Time AI Interview Copilot
A local-first desktop copilot that transcribes live interviews in real time, detects questions, and streams grounded answer suggestions to a private, screen-share-proof cue card.
View source on GitHub- Real-time
- Transcription & question detection
- Local-first
- On-device RAG, encrypted keys
- 3
- Platforms: Windows · macOS · Linux
Overview
BrainCue is a cross-platform desktop app that quietly assists a candidate during a live interview. It listens to system audio, detects when a question is asked, and streams a suggested answer to a floating cue card that only the candidate can see — excluded from screen sharing and recordings. Answers are grounded in the user's own resume, the job description, and personal notes rather than generic boilerplate, and everything runs on-device with the user's own OpenAI key kept encrypted in the OS keychain.
The challenge
Interviews are high-pressure and easy to blank on, but naive AI assistants give generic answers, leak on screen shares, and raise obvious privacy concerns. The app had to capture and transcribe live audio, decide in real time what was actually a question, retrieve the candidate's relevant context, and surface a concise answer fast enough to be useful — all locally, privately, and invisibly to the interviewer.
Our solution
We built an Electron + React + TypeScript app with a multi-window architecture: a main process that owns all sensitive work (OpenAI calls, database, secrets), a renderer for setup and coaching, and an always-on-top overlay window kept out of screen capture. A local RAG pipeline embeds the resume, job description, and notes into a SQLite vector store (Drizzle ORM, better-sqlite3), so answers are grounded in the candidate's material. OpenAI powers transcription, embeddings, answer generation, and text-to-speech, including a mock-interview mode where an AI interviewer asks questions aloud. Global hotkeys and a typed IPC preload bridge tie it together, with the API key encrypted via the OS keychain and never leaving the main process.
Gallery
