EdgeSignal

EdgeSignal is a real-time, hardware-accelerated crypto market intelligence system demonstrating heterogeneous computing on Snapdragon X Elite. It combines live financial data, quantitative microstructure analytics, ARM64 SIMD, GPU computation, and local generative AI in one observable pipeline. EdgeSignal consumes live trades and Level 2 order-book data from Hyperliquid for ETH, BTC, SOL, HYPE, XRP, DOGE, AVAX, and LINK. A native C++ dispatcher routes each market to a persistent worker pinned to a separate Windows logical CPU, allowing eight markets to be processed concurrently. Each worker calculates market-microstructure signals including Hawkes-process order-flow pressure, VPIN, signed VPIN, Order Flow Imbalance (OFI), book imbalance, returns, volume, volatility, and spread. Rolling Z-scores identify abnormal behavior relative to recent market conditions. ARM64 NEON SIMD then processes multiple normalized features simultaneously to calculate direction and activity scores. A deterministic detector converts these measurements into BUY, SELL, or HOLD signals. The LLM does not make the trading decision. The Qualcomm Adreno GPU performs parallel cross-market numerical analysis over a rolling multi-coin feature matrix using native Direct3D 12 compute. It derives aggregate direction, activity, dispersion, signal breadth, and cross-market regime information. For AI interpretation, Qwen3 runs locally through GenieX on the Qualcomm NPU. When a meaningful signal transition occurs, such as HOLD→BUY or HOLD→SELL, the NPU generates a concise interpretation such as BUY_PRESSURE, SELL_PRESSURE, BALANCED, or LIQUIDITY_STRESS. A live dashboard exposes the entire flow: Hyperliquid → CPU Dispatcher → ARM64 NEON SIMD → Adreno GPU → BUY/SELL/HOLD → Qwen3/GenieX NPU EdgeSignal demonstrates a broader edge-AI architecture: CPU for streaming and orchestration, SIMD for low-latency vector math, GPU for parallel numerical analytics, and NPU for local generative AI.