🤖 RegimeLab

A regime-aware, self-learning algorithmic trading bot built with Claude Code agents. Built by a developer who actually runs it — not just backtests it.

---

🧠 The core idea

Most trading bots use the same parameters in every market condition. RegimeLab doesn't.

It detects whether the market is in a bull, neutral, or bear regime — and switches to separately optimized parameters for each. It learns from every trade, blocks patterns that consistently lose, and updates its own skill files automatically.

---

⚙️ Architecture

graph TD
    A([🎯 trading-architect — orchestrator])

    A --> B1
    A --> B2
    A --> B3

    subgraph B1 [" 📊 Market Analysis "]
        direction TB
        B1a[Regime Detector<br/>BULL · NEUTRAL · BEAR]
        B1b[News Sentiment<br/>market awareness]
        B1c[Earnings Calendar<br/>event filter]
    end

    subgraph B2 [" 🛡️ Risk & Learning "]
        direction TB
        B2a[Guardian Agent<br/>monitors + learns]
        B2b[Risk Manager<br/>position sizing]
        B2c[Learning Loop<br/>skill file updates]
    end

    subgraph B3 [" ⚙️ Optimization "]
        direction TB
        B3a[Optuna Engine<br/>regime optimization]
        B3b[Backtesting<br/>walk-forward]
        B3c[Microstructure<br/>slippage + costs]
    end

    B1 --> L
    B2 --> L
    B3 --> L

    L([⚡ Decision Engine<br/>entry · exit · sizing])
    L --> M([🏦 Interactive Brokers])

    style A fill:#7F77DD,color:#fff,stroke:none
    style L fill:#BA7517,color:#fff,stroke:none
    style M fill:#444441,color:#fff,stroke:none

Loading

10 specialized agents working together:

• 🎯 Regime detection — BULL / NEUTRAL / BEAR / TURBULENT
• 🛡️ Guardian agent — monitors behavior, writes learnings to skill files
• 📊 Optuna optimization — separate parameter studies per regime
• 🔄 Learning loop — every trade feeds back into the decision engine
• 📰 News sentiment + earnings calendar awareness
• ✅ Walk-forward validation — real out-of-sample testing

---

📊 Current status

Metric	Value
Mode	Paper trading
Broker	Interactive Brokers (IBKR)
Regimes	BULL / NEUTRAL / BEAR / TURBULENT
Agents	10 specialized Claude Code agents
Optimization	Regime-specific Optuna studies
Infrastructure	Self-hosted Proxmox homelab

---

📁 What's in this repo

This public repo contains the architecture and concepts — not the live strategy parameters or Optuna results. Those are reserved for members.

Folder	Contents
.claude/agents/	Agent definitions (structure, not full logic)
docs/	Architecture diagrams
examples/	Sample backtest output

---

🔒 Members access

Monthly updates with the real stuff:

• ✅ Full agent + skill files
• ✅ Live paper trading results
• ✅ Regime-specific Optuna parameters
• ✅ What worked, what didn't, and why
• ✅ Discord community access

→ Join the waitlist at regimelab.dev

---

🛠️ Stack

Component	Technology
Language	Python 3.11
Database	TimescaleDB
Broker	Interactive Brokers via ib_insync
Agents	Claude Code (Anthropic)
Optimization	Optuna
Infrastructure	Proxmox homelab, Docker, LXC

---

⚠️ Disclaimer

Educational content only. Not financial advice. Past performance does not guarantee future results.