All work
Professional Work · 2025

Omnisound

Ultra-low-latency AI shadowing prototype that listens to user audio and provides instant factual answers while staying silent during casual conversation.

Commercial or client work — public links are not available.

Overview

Omnisound is a real-time conversational AI wrapper designed for 'Shadowing' use cases. Powered by ElevenLabs Conversational AI for near-instant voice interaction, it acts as a passive knowledge engine that intelligently filters out casual conversation, greetings, and background noise — only responding to direct factual queries (stocks, news, weather, math). Integrated with Tavily Search API for real-time data retrieval with strict anti-hallucination protocols. Features a Silence Protocol that ensures maximum 5-word responses, live transcripts via Streamlit, and automatic query classification between noise, subjective questions, and factual queries.

Architecture

ElevenLabs
Whisper
Classifier
Silence Protocol
Tavily Search
5-word cap

Stack

ElevenLabsTavilyStreamlitReal-time