Jarvis V2 — AI Voice Assistant

A voice assistant that only recognizes a fixed command list breaks the moment a user phrases something slightly differently or asks for an app that isn't installed.

Trained a Stacking classifier to predict intent from voice commands, and built a fallback path so anything the classifier can't confidently resolve locally still resolves to something useful — a browser search or web equivalent of the requested action.

1. 01Speech-to-text captures the voice command and normalizes it to text
2. 02A trained Stacking classifier predicts the intended action from the transcribed command
3. 03Matched intents trigger local actions (opening apps, fetching system info)
4. 04Unmatched or low-confidence commands fall back to an equivalent browser action
5. 05Flask backend serves a Gemini-themed web interface for text-to-speech feedback and interaction

• ▸Trained the Stacking intent classifier on a labeled command dataset
• ▸Built the app-launch handler with a browser fallback for commands with no local match
• ▸Wired speech-to-text input and text-to-speech feedback into a single Flask-served interface

• ▸Covering enough intent variety in training data for natural phrasing without over-fitting to exact command wording
• ▸Deciding when to fall back to the browser versus reporting a command as unrecognized

A working voice assistant with local intent prediction and a browser fallback, published as an open-source project.