NEO AI — Your Offline AI PC Assistant

Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady Submission 🦥

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

I built NEO AI, an offline-first Windows PC assistant designed to make everyday computer tasks faster and easier through voice commands.

Instead of relying on a cloud AI model for every action, NEO AI routes supported PC commands directly to local handlers. Say “Open Notepad,” “Increase volume,” or “Take a screenshot,” and the assistant can execute the corresponding action without waiting for an LLM response.

🎙️ Offline speech recognition using Faster-Whisper

🧠 Local conversational AI powered by Ollama

⚡ Direct voice-command routing for supported PC actions

🔊 Offline text-to-speech using Windows SAPI5

🛡️ Safety checks and confirmation for destructive operations

🖥️ A desktop interface with voice controls, activity history, and settings

The goal is simple: less clicking, less waiting, and more control over my own computer.

GitHub repository: https://github.com/gunmasterg9/Neo-AI

Built with Python, PySide6, Faster-Whisper, Ollama, and Windows-native speech synthesis.

I built NEO AI around a local-first architecture using open-source tools and locally running AI models.

Microphone → Faster-Whisper → Local Command Router → PC Action

Text or Speech → Local Ollama Model → Response → Offline Text-to-Speech

The important design decision was separating voice commands from AI conversations.

A straightforward command such as “Open Notepad” should not require an LLM to reason about it. A supported command goes directly to its local handler, while questions and open-ended requests can use the local Ollama model.

Python for application logic and automation

PySide6 for the Windows desktop interface

Faster-Whisper for local speech recognition

Qwen 3 (1.7B) as the configured local model

Windows SAPI5 / pyttsx3 for offline speech output

I also worked on audio resampling, transcription validation, command routing, cancellation, and safeguards against accidental actions from incorrectly recognized speech.

Open innovation made it possible to experiment with a personal AI assistant without making a cloud API the foundation of every interaction.

Using open-source speech tools and a locally running model gives me more control over privacy, latency, customization, and how the assistant behaves.

It also makes experimentation more accessible: I can inspect the pipeline, change the routing logic, test failure cases, and improve the application without depending entirely on a closed AI service.

For me, open innovation is about more than accessing a model. It is about being able to understand, modify, and build on the technology

The idea is to replace repetitive mouse-and-keyboard interactions with short voice commands.

Instead of manually navigating menus, I can ask NEO AI to open a supported application, change volume, or capture a screenshot.

Direct local routing also avoids unnecessary LLM processing for simple commands, making the experience potentially faster and allowing these actions to work even when Ollama is unavailable.

The next milestone is validating the full voice workflow consistently with real microphone input across English, Hindi, and Gujarati.

NEO AI is my exploration of what a useful, privacy-conscious, offline-first personal assistant can become.