FXSentinel — Institutional AI Trading Intelligence Terminal

Retail traders operate at an asymmetric disadvantage: fragmented data, no macro context, no institutional order-flow visibility, and trade execution flows that bear no relationship to the signals they consume. The result is a cycle of late entries, over-leveraged positions, and emotional liquidation.

FXSentinel collapses sentiment aggregation, multi-timeframe structure mapping, volatility filtering, macro event windows, and broker-grade execution into one explainable terminal. Every signal carries its full factor breakdown — no black-box predictions, no recycled retail indicators.

• › Ingest layer — Frankfurter (ECB rates), TwelveData, Yahoo, COT positioning, news polarity.
• › Engine layer — deterministic factor model (currency strength, ATR-normalized risk, structure score, divergence, session bias) → aggregate score + confidence.
• › Synthesis layer — Google Gemini 2.5 Flash translates structured factor outputs into a human rationale. It cannot override direction.
• › Distribution layer — Telegram fan-out, Postgres signals table, email digest, in-app feed.
• › Execution layer — MT5 EA polling, OANDA REST, cTrader OAuth (beta), server-enforced risk gates.

• › React 19 + TanStack Start v1
• › Vite 7 / Cloudflare Workers SSR
• › Tailwind v4 + shadcn/ui
• › Supabase (Postgres + RLS)
• › Lovable AI Gateway (Gemini, GPT-5)
• › pg_cron + pg_net schedulers
• › MT5 Expert Advisor (MQL5)
• › Stripe billing (live + sandbox)

The engine is stateless per-run; pg_cron handles fan-out at 1-, 5-, and 15-minute cadences. Signal storage is Postgres with row-level security; broadcast is Telegram + WebSocket. Worker-based SSR scales horizontally with no cold start; the EA bridge runs on the user's machine, so execution scales linearly with users at zero marginal cost.