Offline AI Voice Assistant Build
Budget: ₹750 – ₹1,250 INR
I’m putting together an entirely offline voice assistant named “Aizen.” Everything must run locally on Windows and ship as a single-file .exe produced with PyInstaller, so no cloud calls are allowed at any point.
Core stack
• Vosk handles real-time speech-to-text.
• LLaMA 3 (quantized) interprets intent and decides what to do next.
• pyttsx3 provides the reply voice, and I’ll want room to fine-tune rate, pitch, and voice persona.
What it has to do on day one
— Wake, listen continuously, and transcribe in near real time.
— Recognise intents such as “open browser,” “search for…,” and simple typing instructions.
— Open Google Chrome automatically when requested, then pass the search or typed text straight into the active field so a user sees immediate results.
Architecture expectations
I need clear, well-commented modules for STT, NLU, TTS, and action execution so I can bolt on new skills later without touching core logic. Dependency management, model downloads, and configuration should all be scripted to make first-time setup painless.
Deliverables
1. Source code with README that explains installation, model paths, and how to add a new intent.
2. PyInstaller spec and resulting Windows .exe that runs fully offline.
3. Basic test script or instructions that prove the assistant opens Chrome and performs a spoken search query end-to-end.
If everything works smoothly I’ll roll the same framework out to additional apps and browsers, so writing clean, extendable code will be valued just as highly as the immediate features above.
Core stack
• Vosk handles real-time speech-to-text.
• LLaMA 3 (quantized) interprets intent and decides what to do next.
• pyttsx3 provides the reply voice, and I’ll want room to fine-tune rate, pitch, and voice persona.
What it has to do on day one
— Wake, listen continuously, and transcribe in near real time.
— Recognise intents such as “open browser,” “search for…,” and simple typing instructions.
— Open Google Chrome automatically when requested, then pass the search or typed text straight into the active field so a user sees immediate results.
Architecture expectations
I need clear, well-commented modules for STT, NLU, TTS, and action execution so I can bolt on new skills later without touching core logic. Dependency management, model downloads, and configuration should all be scripted to make first-time setup painless.
Deliverables
1. Source code with README that explains installation, model paths, and how to add a new intent.
2. PyInstaller spec and resulting Windows .exe that runs fully offline.
3. Basic test script or instructions that prove the assistant opens Chrome and performs a spoken search query end-to-end.
If everything works smoothly I’ll roll the same framework out to additional apps and browsers, so writing clean, extendable code will be valued just as highly as the immediate features above.