AI Desktop Assistant Python project (Kukorica)


I developed a modular, voice-driven AI desktop assistant in Python that bridges LLM intelligence with real Windows-level automation.


Core Architecture


I combined GPT-4o-mini / o4-mini with offline XTTS voice cloning and a custom priority-based IntentMatcher system supporting multilingual (HU/EN) execution.


Hybrid Intelligence Layer

  • Local intent recognition (HU/EN keyword-based rules)
  • Priority-based intent matching
  • Low-latency execution for system commands
  • LLM fallback for complex interpretation

Voice Processing Pipeline

  • Speech-to-text
  • Local intent resolution
  • LLM interpretation (when required)
  • Tool execution layer
  • Offline XTTS voice response

Desktop Automation

  • Open and close applications
  • Shutdown and restart computer
  • Close active or all windows
  • Open Windows settings
  • Retrieve system time
  • Open folders
  • Open browser URLs

Article Processing

  • Article detection and reading automation
  • Fast-path processing for Hungarian news sites
  • Telex support
  • Index support

Developer Features

  • AI-based refactoring of selected code
  • Modular controller architecture
  • Tool-based execution mapping

Screen Capture Module

  • Voice-triggered screenshot creation
  • Reopen last screenshot
  • Capture storage management

Communication Layer

  • Send SMS
  • Make voice calls
  • Send Viber messages

Philosophy


The assistant was designed as a real operating system companion, not just a chatbot interface. It connected AI reasoning, voice interaction, and direct Windows automation into a unified desktop control system.

MEDIA




CONTACT


info [AT] kozari.eu
KOZARI Software