Back to projects

// PROJECT — AI

Crypto Trade Agent

Solo design, build, and deployment

Most crypto signal tools stop at displaying ideas. Crypto Trade Agent goes further: every setup is automatically validated against live Binance market data and pushed through a server-side risk engine before it can become a (paper) trade. I built the whole system — schema, risk rules, execution loop, and dashboard.

The architecture deliberately separates signal generation from execution, the same separation-of-duties principle you'd apply to any system where one component shouldn't be able to act unilaterally. A Vercel Cron heartbeat fires every minute to drive the ingestion-validation-simulation cycle, with duplicate-prevention guards stress-tested under bursty cron firing so noise never pollutes the trade log. Trades, configurations, and run telemetry live in Supabase behind row-level security, with explicit metadata like market_regime and setup_strength_score recorded on every run so strategies can be analysed — and re-weighted — from real telemetry rather than vibes.

// Built With

Next.jsTypeScriptSupabaseBinance APIVercel CronTailwind CSS

// Highlights

  • Separation of signal generation from execution — a server-side risk engine validates every trade before it reaches the simulated order book.
  • One-minute Vercel Cron heartbeat drives the ingestion → validation → simulation loop, with duplicate-prevention guards stress-tested under bursty firing.
  • Telemetry-first schema: market_regime, setup_strength_score, and full run metadata recorded in Supabase so strategy weights are tuned from real paper-trade results.
  • Soft-reset workflow preserves historical telemetry across reseeds — strategy analysis survives data resets.
  • Premium live dashboard with dynamic trading modes built on Next.js and Tailwind.

Security Considerations

  • All exchange API keys live server-side in Vercel environment variables — nothing sensitive ships to the client.
  • Supabase row-level security on trades, configuration, and telemetry tables.
  • Strict 404 behaviour enforced on unknown routes to keep the API surface minimal and unenumerable.
  • Anti-spam guardrails rate-limit the execution loop so a misfiring signal source can't flood the trade log.

Project Timeline

Discovery / Goal

Mar 2026

Identified a need for a crypto signal terminal that doesn't just display ideas but automatically validates and paper-trades them. The goal was a secure, anti-spam execution loop that simulates real market conditions end-to-end.

Design Decisions

Mar 2026

Separated signal generation from execution. Built a server-side risk engine to validate every trade and used Vercel Cron to fire one-minute heartbeat cycles, all while keeping a premium dashboard aesthetic.

Schema & Persistence

Mar 2026

Modelled trades, configurations, and run telemetry in Supabase with row-level security. Added explicit metadata such as `market_regime` and `setup_strength_score` to make later strategy analysis tractable.

Implementation

Mar 2026

Built the dashboard with dynamic trading modes, layered the risk management rules, and constructed `run-bot` endpoints for signal ingestion, validation, and order simulation.

Testing & QA

Apr 2026

Verified the build locally, enforced strict 404 behaviour on unknown routes, kept all API keys server-side via Vercel environment variables, and stress-tested duplicate-prevention limits under bursty cron firing.

Deployment

Apr 2026

Pushed to Vercel production with environment variables synced, Cron jobs registered, and Supabase migrations applied through the deploy pipeline.

Iterations

Apr 2026

Added a soft-reset workflow to preserve telemetry across reseeds and refined the strength-scoring weights based on the first weeks of paper-trade results.