A forex day trading system with AI embedded at every layer — deciding what to trade, when to trade, and how much to risk, then handing execution to IG's server-side infrastructure — building confidence from multiple independent signals before risking any capital.
👁️ The Observation
84% of retail forex traders lose money. Social Media has polluted the credibility of day trading. Get rich quick schemes which result in losses and uneducated financial risk, continue to plague hard working people.
Making trading available to people but risk adverse, and having confidence on a platform to make the right trade at the right time, is whats missing. This is the gap confluence plugs.
⚙️ How It Works
The whole system runs in a small set of docker containers, always on, scanning 10 forex pairs every 5 minutes. Each scan pulls data from over 20 sources: technical signals, macro context, sentiment, institutional positioning, and feeds it all into a confidence score formula. A 7-signal Market Activity Detector (including TradingView technical analysis and CNN Fear & Greed Index) gates whether the market is worth trading at all. Session Guard restricts activity to data-driven profitable hours. The bot decides WHAT to trade, WHEN to trade, and HOW MUCH to risk. Once a trade is placed with an IG native trailing stop, IG manages the position server-side with zero latency — no polling, no local process keeping the trade alive. Every score, every skip, every trade is logged. So I can tell you why any trade happened, and why most of the time, it doesn't.
%%{init: {'theme':'base','themeVariables':{
'fontFamily':'-apple-system, BlinkMacSystemFont, Segoe UI, Roboto, Inter, sans-serif',
'fontSize':'15px',
'primaryColor':'#ffffff',
'primaryTextColor':'#0f172a',
'primaryBorderColor':'#fbcfe8',
'lineColor':'#db2777',
'edgeLabelBackground':'#ffffff',
'clusterBkg':'#ffffff'
}}}%%
flowchart LR
A([10 forex pairs<br/>scanned every 5 min]) --> B[Technicals<br/>MACD · RSI · EMA<br/>Bollinger · ATR]
A --> C[Market context<br/>20+ external feeds]
A --> D[ML layer<br/>shadow mode]
B --> E{Confluence<br/>score}
C --> E
D --> E
E -->|≥ 70% agree| F([Trade placed<br/>IG trailing stop])
E -->|mixed| G[Wait and log]
style A fill:#fff1f2,stroke:#e11d48,stroke-width:2px,color:#9f1239
style B fill:#ffffff,stroke:#fbcfe8,stroke-width:1.5px,color:#0f172a
style C fill:#ffffff,stroke:#fbcfe8,stroke-width:1.5px,color:#0f172a
style D fill:#ffffff,stroke:#fbcfe8,stroke-width:1.5px,color:#0f172a
style E fill:#fdf2f8,stroke:#db2777,stroke-width:2.5px,color:#9d174d
style F fill:#e11d48,stroke:#e11d48,stroke-width:2px,color:#ffffff
style G fill:#f8fafc,stroke:#cbd5e1,stroke-width:1.5px,color:#64748b
linkStyle default stroke:#db2777,stroke-width:1.5px
🎯 The Confluence Principle
Confluence is where independent things come together. Rivers meeting, roads crossing, signals agreeing. In trading it means the same idea showing up in multiple places at once. Momentum says buy, trend agrees, higher timeframe confirms, context doesn't argue. One signal on its own doesn't mean much. Five or more pointing the same way is a different thing entirely. That's what this system is built to find, and what gets a trade placed.
Momentum · 40%
MACD crossover confirmed by RSI not being stretched
Trend · 25%
EMA alignment across 20 and 50 period moving averages
Mean-Reversion · 20%
Bollinger Band position: overextended or trending?
A trade only fires when the weighted score clears 70%. Most of the time that means waiting. The point isn't big days, it's boring ones repeated. Steady, reliable profits over months, not lucky runs that get wiped out in a week. Every pound risked has earned the right to be risked. Integrity not vanity. De-risking capital, not chasing it.
✨ Why This Is Different
Most forex bots trade on one signal and a prayer. This one won't place a trade until the evidence stacks up across five technical readings, twenty external context feeds, a Market Activity check, and a higher timeframe confirmation. No single indicator gets to decide. AI isn't bolted on — it's woven through signal generation, execution, risk management, and self-improvement. That's the core difference.
✕ Single indicator triggers
✓ 5 technical + 20 context signals must converge
✕ Black-box neural networks
✓ Fully explainable formula, every score logged
✕ ML trained on price alone
✓ LSTM gated to 85% accuracy before live weighting
✕ Client-side stop management, polling delays
✓ IG server-side trailing stop, zero latency, ATR-scaled entry
✕ "Trust me it works"
✓ Full dashboard with SQLite audit trail
🧠 AI at Every Layer
AI isn't a single feature — it's the operating system. Claude AI and machine learning are embedded across every layer of the platform, from generating signals to closing trades to learning from mistakes.
Signal
Claude AI confidence formula scores every pair on every scan. RSI exhaustion filter prevents entries into overbought/oversold reversals. Every score is explainable.
Execution
IG native trailing stop manages every position server-side after entry — zero latency, no local polling. The bot decides WHAT, WHEN, and HOW MUCH. IG handles trade management.
Intelligence
7-signal Market Activity Detector combines TradingView technical analysis, CNN Fear & Greed Index, and 5 internal signals to gate whether the market is worth trading. LSTM neural network runs in shadow mode, gated to accuracy before live weighting.
Operations
38 silent failure monitors. Automated churn detection spots when the bot is entering and exiting too frequently. Hourly P&L checks. Daily health audits. All anomalies pushed to Telegram.
Security
OWASP compliance skill scans every code change. Code drift detection. Config validation on boot. Cloudflare Access with Google SSO.
Claude Code as pair programmer with 18 specialist skills: OWASP security, backtesting, regime detection, volatility modelling, walk-forward validation, and more. Full DevOps deployment discipline on every push.
📊 The Evidence
Six weeks of building, learning, and iterating. 150+ merged PRs, 500+ commits, 20+ external data feeds, and 18 Claude Code skills installed. The early numbers are pointing in the right direction and everything below is auditable, logged, and published. See for yourself.
Independent signals required to converge on every trade
Those 14 signals get weighted, scored, and AI reviewed for thesis, risk, and context. A trade only fires when the confidence score clears an agreed number. At present that is 70% and above.
🔌 IG API Integration
The system is built on deep, production-grade integration with the IG Group API. Not a wrapper around a single endpoint — a full-stack implementation covering auth, data, execution, and streaming.
REST API
Full authentication flow (CST + X-SECURITY-TOKEN with 6h auto-refresh), position management, order placement, candle retrieval, account switching between demo and live.
Lightstreamer
TRADE subscriptions for position updates and close detection (trailing stop vs stop loss vs take profit). MARKET subscriptions for real-time price streaming across all 10 pairs.
Position Sizing
Pip-based sizing for both CFD and spread betting account types. ATR-scaled stop loss calculation. 2% risk-per-trade model with affordability checks.
Data Caching
10,000 data points/week budget managed via in-memory cache with TTL matching timeframe duration. Tops up with 3 candles instead of re-fetching 60. Actual usage: ~1,320 points/week.
Epic Mapping
10 forex pairs mapped to IG epic codes. Extensible architecture ready for indices, commodities, and crypto epics.
Demo to Live
Migration path from demo to live account planned for end of May 2026. Spread betting account type for tax-free gains in the UK. Same codebase, account switch via config.
🗺️ Roadmap
Six weeks from zero lines of code to a live demo account bot with full auditing, 20+ external data feeds, 18 Claude Code skills, and a secured GitHub repo ready for governed collaboration. That's where we are. Here's what's next.
Live Now · Secured Infrastructure
Built on private infrastructure. Every commit tracked. Every deploy audited.
Private GitHub Repo
Cloudflare Tunnel + Access
Google SSO
Docker Containers
CI/CD on Every Push
Anthropic Claude AI
SQLite Audit Trail
IG Group API
Lightstreamer Streaming
IG Native Trailing Stops
18 Claude Code Skills
OWASP Compliance
📝 500+ commits · 🔀 150+ merged PRs · 🔍 Full audit trail on every trade decision
Now
Q2 2026
Hardening
Market Activity Detector live
Session Guard active
IG native trailing stops live
Autonomous tuning design
Next
Q3 2026
Go live
Spread betting migration
Live trading (end of May)
Autonomous tuning engine
Later
Q4+ 2026
Scale
Multi-asset (indices · commodities)
Cloud hosting for 24/7 uptime
Published IG Group Claude Code skill
💬 We're Looking For Three Conversations
01 · Day traders
Critique the model
Tell me what I've got wrong. Signal weighting, weekend risk, how I handle correlation. Pick it apart.
02 · Engineers
Build with us
If you like the problem, come build with me. Multi asset work, ML pipeline, broker abstraction. Open source.
03 · Capital partners
Post-review only
Not yet. Once traders have properly torn this apart, we can talk. Ring fenced capital, measured live deployment.
🤝 AI-embedded trading. Built on IG. Tell me what I've missed.
Every review makes this sharper before any real money goes in. No pitch deck, no sales call. Just your honest critique from someone who actually trades for a living.
Reward£50 Amazon Voucher or equivalent Charity Donation for any review that leads to a measurable and profitable improvement.