Back to Projects
BARQ — Voice-Controlled AI Desktop Assistant
A voice-first AI desktop assistant combining wake word detection, natural language understanding, and automation in one cross-platform app. Control apps, search jobs, create content, and more — all by voice, like Alexa for your computer.
7
Technologies used
3
Features implemented
1
Challenges solved
Open Source
Status
Tech Stack
Frontend
React
Backend
PythonFastAPI
Database
SQLite
Impact
Production-grade desktop app with local LLM inference, streaming voice responses, and self-updating installers for macOS and Windows.
Implementation Details
- BARQ pairs a Python FastAPI backend with an Electron + React desktop UI.
- Always-on Vosk wake word detection triggers a Deepgram Voice Agent over a single WebSocket for hands-free conversation, while a local Ollama LLM handles chat with transparent fallback to OpenAI-compatible cloud APIs.
- It ships with multi-brain knowledge graphs (domain-specific NetworkX graphs with auto-extraction and a 3D neural dashboard), job search automation across major boards with ATS-optimized matching, desktop automation workflows, and a one-click installer that self-updates from GitHub Releases.
Challenges & Solutions
- Voice latency was the hardest constraint — the app streams sentence-aware TTS so speech starts while the LLM is still generating, and routes voice endpoints locally while sending heavier LLM calls to a remote Oracle VM backend to keep the desktop app responsive.