Testnet
2026Self-reported, private repository. Testnet only; no real money. The case study lists its own caveats.

Regime-Switching Scalper

Multi-asset scalper with runtime market regime detection (testnet)

Media & Demos

Role

Developer

Period

2026

Category

trading

Overview

A multi-asset scalper that detects the market regime at runtime and switches to the specialist configuration chosen for that regime. It runs on Binance testnet, with no real money, and is collecting a baseline before any decision about live capital. It reads direct Binance WebSocket streams instead of CCXT and buffers klines instead of polling REST. Configurations were chosen with walk-forward validation on historical data; the figures are self-reported.

Key Highlights

  • Runtime regime detection (Trending / Ranging / Volatile) per asset per tick
  • Direct Binance API WebSocket streams — CCXT eliminated for lower overhead
  • Lower latency from direct exchange connectivity (self-reported measurement)
  • WebSocket kline buffer replaces REST polling (cycle time 7,215 ms to 691 ms, self-reported)
  • Specialist config per regime, selected with walk-forward validation on historical data
  • Runs continuously on testnet across multiple assets; no real money
  • Distributed walk-forward backtest on a Ray cluster across three cloud VMs

Tech Stack

Python
FastAPI (Async)
WebSockets
Redis
Walk-Forward Validation
HMM / Regime Detection
Direct Binance API

What I Learned

The main insight was that no single parameter set suits every market regime. Classifying the market state at runtime and switching configs avoids the fragility of one over-fitted static setup. Walk-forward validation was non-negotiable: in-sample performance means little without out-of-sample confirmation. Out-of-sample results on historical data are not evidence of future profit.