VoxCall Oracle: Live Audio Trading Agent

The latency gap in modern finance is a multi-million dollar problem. Traditional algorithmic trading relies on text-based transcripts that humans or bots read minutes after words are actually spoken on an earnings call. By the time the data is digested, the market has already moved. VoxCall Oracle is an end-to-end, production-ready pipeline designed to bridge the gap between raw audio and instant market execution. Our fully autonomous agent listens to live earnings calls and executes trades in real-time without human intervention. We utilized four core technologies to build this stack: 1. Speechmatics: We use their best-in-class speech-to-text API for real-time transcription and speaker diarization. This allows our agent to know exactly *what* was said and *who* said it (e.g., separating a confident CEO from a cautious reporter). 2. Featherless AI: To eliminate the risk of AI hallucinations, we route the transcript chunks into a 3-model financial ensemble via Featherless (Llama-Open-Finance, Fin-o1, and finance-chat). These specialized models vote on the market sentiment to ensure pinpoint accuracy. 3. LangGraph: In automated trading, safety is critical. We built a LangGraph state-machine that acts as a risk-gating firewall. Trades are only approved if the AI ensemble's confidence score exceeds a strict user-defined threshold (e.g., >75%). 4. Kraken CLI: Approved signals are instantly routed to the Kraken API for ultra-low-latency order execution. Our entire platform is wrapped in a premium, glassmorphism UI deployed on Streamlit Cloud, featuring a live Paper PnL tracker and execution logs. VoxCall Oracle isn't just a script—it's a fully functional, cloud-native hedge fund analyst.