How We Build Our Trading Bot in the Open

There's a lot of noise in the crypto trading space. Promises of guaranteed returns. "Set it and forget it" bots. Signals groups selling dreams. We wanted to be different.

So we're doing something unusual: we're building our trading bot in public. Not the hype kind of "building in public" — not screenshots of glowing dashboards or cherry-picked P&L screenshots. We're sharing the real deal: the code, the strategies, the decisions, and yes, the losses too.

What the Bot Actually Does

Our trading bot operates in two markets: cryptocurrency and Polymarket, the prediction market platform. It's currently running in paper trading mode — meaning every trade it makes is simulated with $10,000 in virtual capital. No real money moves. This is by design.

We believe in earning the right to trade with real capital. Before we ever flip the switch to live trading, we need to prove the system works over time. That's the dry_run principle at the core of everything we do.

The Tech Stack

The bot is built in Python, running on an asyncio loop that polls market data every 15 minutes. Here's what makes it tick:

• 9 Trading Strategies — From mean reversion and RSI-based swing trading to MACD crossovers, Bollinger Bounce, volume breakouts, DCA laddering, and minute scalping. Each strategy has specific conditions under which it activates.
• LLM-Augmented Decisions — The bot doesn't just follow rules. It uses large language models (specifically Qwen3.5-plus via DashScope) to analyze market regimes, evaluate trade setups, and make nuanced decisions about whether to buy, sell, hold, or adjust risk.
• Confidence-Based Execution — Every decision comes with a confidence score. High-confidence trades (85%+) can execute automatically. Lower-confidence trades get queued for human review via Telegram.
• Regime Detection — The bot constantly monitors market conditions. When volatility spikes or trends shift, it automatically dials down position sizes and deprioritizes weaker strategies.

Safety First, Always

The most important thing about our system isn't what it does — it's what it won't do. We've built multiple layers of protection:

• dry_run = true — This isn't just a config flag. It's baked into the code at multiple levels. Even if someone changed the config, the execution layer has its own hardcoded guard.
• Position Limits — Maximum 5% of portfolio per trade, 7% total exposure per position. Hard caps prevent over-leveraging.
• Daily Budget Caps — $1.50 USD maximum in simulated trading costs per day. We track every simulated fee.
• Circuit Breakers — After 3 consecutive losses, the bot automatically pauses and requires human review. Same for 4% drawdown or 3% daily drawdown.
• Prompt Injection Guards — The LLM receives a sanitized system prompt. We actively block attempts to inject malicious code through the prompt itself.

What We've Learned So Far

Building this system has been humbling. Here's what stands out:

Regime matters more than strategy. A strategy that works beautifully in a ranging market gets crushed in a breakout. We've learned to build "regime awareness" into the core decision loop — the bot now weights strategies differently based on current market conditions.

LLMs are great at reasoning, not prediction. The language model doesn't "know" what the price will do. But it's excellent at checking its own reasoning, identifying contradictions in its analysis, and flagging uncertainty. That's where the real value is.

Human oversight isn't optional. We've seen the bot make what seemed like solid decisions that turned out badly. The confidence threshold system — requiring human approval for anything below 85% confidence — has prevented several potential drawdowns.

Why Build in Public?

Transparency builds trust. In an industry full of opaque "black box" systems and unverifiable claims, we wanted to show exactly how our system works.

We're also holding ourselves accountable. When you build in public, you can't hide failures. Every loss is logged, every decision is recorded. This discipline makes us better traders.

And honestly? We hope this inspires others. The trading bot space doesn't need more hype. It needs more transparency, more open source collaboration, and more honest conversations about what works and what doesn't.

What's Next

We're not rushing to live trading. The goal is simple: prove the system works over 6-12 months of paper trading. Only then will we consider transitioning to real capital — and even then, with position limits and circuit breakers firmly in place.

We'll keep sharing updates. The decisions log is public. The insights are logged. When we win, you'll see it. When we lose, you'll see that too.

That's the point. Building in the open, one trade at a time.