TrumpTrader

From tweet to position, automatically.

The premise is pure speed: a webhook catches a Trump tweet the moment it posts, Grok 4 scores its market impact into a structured decision — signal, confidence, expected magnitude — and if the confidence clears the bar, the engine places a BTC perp trade on AsterDEX with staged take-profits and a stop-loss.

The system is multi-account: several Twitter accounts can be monitored in parallel, each with its own prompts, confidence multipliers and risk settings, all managed from an admin UI.

The prompt and the guardrails.

Two parts made this work: getting structured judgment out of the model, and never trusting it blindly.

• The Grok system prompt is a structured-output exercise — a factor catalog (crypto mentions, Fed policy, tariffs), a confidence rubric, and seven few-shot examples that force clean JSON decisions.
• The trading engine wraps every trade in safety gates: a global kill switch, confidence threshold, single-position lock, cooldown timer, daily trade limit and liquidation warnings.
• AsterDEX has no off-the-shelf SDK, so I implemented its Web3 auth by hand — ABI-encoding and signing requests with separate main and API wallets.

A demonstrator, not a money printer.

This was a fast experiment in event-driven trading, not a fund. The README says it plainly: at high leverage a 1% move against you is liquidation. The interesting part is the pipeline — real-time ingestion, LLM judgment, gated execution — not a promise of returns.