-
-
Newly updated name and dashboard, more appealing looks and design
-
backend live on render hosting services
-
The bot charges 15% of the user monthly profit, 30 days free trial to convince the users
-
Gemini API used as an external bot advisory to validate signal confidence.
-
-
.
-
-
-
-
-
paparrazi
Declaration
This project has a previous foundation. I originally built it as a personal XAUUSD scalping bot, and for this submission I iterated on that base to make it a multi-pair trading system with multi-user support, subscription billing, and a more scalable cloud architecture. The current version includes significant new work beyond the original bot: I built a mobile app for easy user access, multi-account orchestration, broader market scanning, and a redesigned backend for shared use.
Inspiration
I started trading gold with a small account — no stop-loss, no take-profit, just me and the chart for hours. I'd watch every tick, convinced I could exit at the perfect moment. But it only took looking away for a few minutes. One retracement, and the account was gone. Over and over. I realized I needed something I could never be: always watching, never emotional, never distracted. So I built it — an automated system that sits on the chart 24/7, catches pullbacks in the direction of the trend, and never blows up from a retracement I wasn't there to see.
What it does
An always-on algorithmic trading bot that trades gold and US indices (XAUUSD, US30, US100) — the only instruments that survived honest backtests with real costs. It runs on Render cloud infrastructure, executes via Capital.com, and manages risk through equity-scaled position sizing with no hard stop-losses — instead trailing a giveback fraction of the winning wave and force-closing at a fixed time horizon (the "Pull-into-H1 Scalper"). It supports multiple users, subscription billing, a mobile app, and a leader-lease failover architecture so trading survives crashes.
How we built it
Python FastAPI backend with a thread-pool bot architecture — each user account gets its own async bot instance. The strategy is an M5 pull-into-H1 scalper: it trades only in the direction of the last completed H1 candle body, enters after a pullback of at least 0.30× ATR against the trend, and trails a giveback fraction of the wave from the actual fill, force-closing after a fixed number of M5 bars. Entries and exits are priced from actual fills, so live == backtest. Risk is managed per-event and per-day with configurable loss caps. Historical data from Dukascopy (~23 years of XAUUSD M1, plus full M1 history for 12 pairs). Paystack and MaxelPay handle subscription payments; a Flutter mobile app gives users live control and monitoring.
Challenges we ran into
Building a stop-loss-free system that doesn't blow accounts required extensive backtesting to separate honest strategies from artifacts. We discovered that our original intra-candle wave scalper was only profitable because backtests filled at ideal "level" prices — under real market fills it was a loser (XAUUSD PF 0.53, −286R), so we removed it and replaced it with the Pull-into-H1 strategy that passed bar-close fills with real costs. Capital.com rate limits and session timeouts demanded a robust reconnection layer with exponential backoff. Synchronizing multiple bots sharing a single broker account while keeping state consistent across restarts was harder than expected — so was proving the live engine behaved identically to the backtest (we built a replay harness that matches trade-for-trade).
Accomplishments that we're proud of
The trailing-giveback exit logic that lets winning runs breathe without giving back everything — backtests show it captures significantly more upside than fixed-TP models on trending days. A proven live == backtest pipeline: the live engine reproduces the validated sweep exactly. An honest validation pipeline where every traded pair was proven profitable out-of-sample: US30 profit factor 1.81, XAUUSD 1.40, US100 1.34 on 2026 OOS data after real costs. Per-pair auto-calibration so the bot tunes its own pullback/trail parameters from recent market structure. ~23 years of M1 data for gold, systematically archived and queriable for any date range.
What we learned
That sophisticated exits matter more than entries — a mediocre entry with a smart exit consistently outperforms a perfect entry with a dumb stop. Equity-scaled sizing is critical for small accounts; fixed lot sizes are a fast path to ruin. Cloud deployment of trading systems is viable if you handle broker disconnections gracefully. And that a validation harness that proves the live engine matches the backtest — under honest fills with real costs — is worth more than a fancy model, because it catches strategies that only look profitable on paper.
What's next for Forex Gold scalping AI
A disciplined expansion of the live instrument set — the 12-pair data and models already exist, but each pair must clear honest OOS validation before it trades real money. Live PnL streaming to the mobile/web dashboard via WebSocket, correlated-pair coordination across more instruments, and a community marketplace for sharing and subscribing to strategy configurations. The business model
- Product: An automated trading bot that trades for you across gold and US indices.
- Free trial for demo accounts (play money) — no limits, always free.
- 30-day free trial for live accounts (real money) — full access, no payment needed.
- After the trial: the bot tracks your profit in 30-day periods. If you made profit, you owe 15% of that profit as a fee. If you lost money, you owe nothing.
- If you don't pay, the bot stops working until you settle — the startup only makes money when users make money. This aligns incentives — the bot needs to be profitable for both the user and the company.
- Demo is free forever — acts as a marketing funnel to get users hooked, then convert them to live accounts.
Log in or sign up for Devpost to join the conversation.