Please explain Angular lifecycle hooks.
Whisper.cpp input - transcription 1.0sRealtimeAssistant
A production-grade live assistant for technical interviews, meetings, and screen-aware workflows. It listens, transcribes, streams answers, captures context, and can run fully local with Qwen and Whisper.cpp when privacy matters.
- Local + Cloud
- Qwen, Whisper.cpp, Apple Speech, GPT Realtime-2
- Live Context
- Mic, system audio, screenshots, recordings
- Practice Mode
- Karaoke readback with highlighted responses
Angular lifecycle hooks are methods called at key points in a component lifecycle: initialize, check, project content, render view, and clean up.
Local Qwen - first token 1.4s - 19.4 tok/sAngular lifecycle hooks help you explain setup, change detection, view updates, and cleanup.
Karaoke readback - Apple Speech
Realtime stackQwen GGUFWhisper.cppGPT Realtime-2Live interview answers
Capture the question from microphone or system audio, then stream a concise answer from GPT Realtime-2 or local Qwen.
Local-first transcription
Use Whisper.cpp, Whisper CLI, Apple Speech, or Qwen ASR depending on whether speed, privacy, or accuracy matters most.
Screen-aware context
Attach screenshots, detected moments, and visible app context so answers can react to the actual work in front of you.
Karaoke readback
Read the latest model response aloud while the app highlights your progress and tracks improvised wording.
Model routing
Choose the right engine for the moment
RealtimeAssistant is built for the messy reality of live speech: short questions, long answers, missed edge words, code terms, accents, background noise, and privacy-sensitive interviews.
Local Qwen GGUF answers with token streaming
Whisper.cpp ASR with VAD chunking and Apple Speech draft support
GPT Realtime-2 path for low-latency cloud sessions
Credit-aware usage tracking for paid production accounts
Workflow
From speech to useful answer without losing context
The app separates capture, ASR, answer generation, and readback so every session can be tuned for latency, accuracy, cost, or local privacy.
- 1
Listen to microphone, system audio, or both selected channels
- 2
Transcribe with the selected ASR path and preserve partial progress
- 3
Combine ASR candidates when Whisper or Apple Speech misses an edge word
- 4
Stream the answer into the session timeline with timing and model labels
Production controls
Built for real usage, not a demo script
Local models are excluded from cloud usage billing, GPT Realtime sessions can be measured with credits, and model labels plus timing metadata make every response traceable.
Private local mode
Run interviews with local Qwen and Whisper.cpp
Use local ASR and local answers when you need the Mac to handle sensitive prompts without sending the session to cloud models.