BravoBot – Voice-Activated Personal Assistant

Project Type: AI Assistant · Voice + Vector Memory + LLM

Overview

BravoBot is a modular, voice-activated AI assistant that runs entirely from the command line. It transcribes user speech, classifies the intent using a fine-tuned transformer model, and performs actions using automation, memory, or generative AI. It is designed to run offline-first with local LLMs (like LLaMA via Ollama), and can optionally fall back to OpenAI when needed.

Motivation

Most voice assistants either rely heavily on cloud APIs or lack extensibility. BravoBot was created to be a fully offline-capable, modular system for power users. With built-in voice UX, persistent semantic memory, and plugin-style routing, it enables personal productivity, automation, and AI interaction — all locally.

Key Features

• Custom wake word detection ("Ben") for privacy and control
• Intent classification using a fine-tuned DistilBERT model
• Offline and online LLM support (Ollama and OpenAI GPT fallback)
• Vector-based semantic memory using FAISS + sentence-transformers
• Memory recall through similarity search and LLM prompt injection
• Voice-controlled commands: Google search, YouTube, time, weather
• Note-taking, summarization, and CSV export via voice
• Session log summarization using AI
• Interruptible TTS using Shift + Q hotkey
• Fully modular plugin routing via context-aware handlers

Architecture & Workflow

Speech is transcribed using Whisper and passed to an intent classifier (DistilBERT). Based on the intent, control is routed to the appropriate plugin handler (e.g. vector memory, automation, LLM query). Semantic memory is stored locally using FAISS; top matches are retrieved based on the user's prompt and injected into the LLM context to enhance responses. All components run locally by default, with optional internet-based LLM fallback.

Technologies Used

Python, Whisper, DistilBERT, Hugging Face Transformers, FAISS, Sentence-Transformers, PyTorch, Ollama, OpenAI GPT, pyttsx3, Selenium, REST APIs

View GitHub Repository

← Back to Portfolio