Innoval Trading // Live Walkthrough
I built an AI quant
trading system from scratch.
Multi-asset. Paper-first. Every number computed by real code, not a chatbot guessing. Here's the whole thing, end to end.
No hype. No fake returns. Just the engineering.
Why
Most "AI trading bots" are toys.
Screenshots, hype, backtests that fall apart the second they hit new data. I wanted the opposite: a system that finds a real edge, controls risk, and tells the truth about what actually works.
× What it's not
No black-box signals. No guaranteed returns. No curve-fit fantasy that only ever worked on the past.
✓ What it is
An edge-finding, risk-controlled machine that validates its own strategies and runs itself, around the clock.
The rules it's built on
Four rules, no exceptions.
▤ Real numbers only
Every metric is computed by code on real data. The AI explains and proposes. It never invents a number.
↷ No look-ahead
A position decided at today's close only earns tomorrow's move. Enforced in code, proven by tests.
⇅ Costs always modelled
Fees and slippage on every single trade. No frictionless fantasy that vanishes in the real world.
🔒 Paper-first, live-gated
It cannot touch real money until I explicitly switch it on, behind hard risk limits and a kill switch.
How it's built
Six layers, one pipeline.
01
Data
Crypto, stocks, FX, macro. One feed.
→02
Engine
Own backtester + real metrics.
→03
Strategies
Rules in, target positions out.
→04
Validation
The gauntlet. PASS or die.
→05
Execution
Paper bot + risk gate + ledger.
→06
Command Centre
Watch it all, live.
Asset-agnostic by design. Crypto, equities, FX and macro all run the exact same path.
The engine
I wrote my own backtester.
No off-the-shelf black box. A vectorized engine I control completely, so I know exactly how every number is produced and trust it when it matters.
Built in
No look-ahead
Tomorrow's move on today's decision. Proven by tests.
Built in
Fees + slippage
Real-world friction on every position change.
Outputs
Real metrics
CAGR, Sharpe, Sortino, max drawdown, win rate.
The honest part
Every strategy runs a gauntlet.
Backtest
Does it work at all?
→Walk-forward
Does it hold on data it never saw?
→Monte Carlo
Is it robust or just lucky?
→Drawdown
Could you actually survive it?
→Verdict
The system decides.
PassConditionalFail
A backtest alone lies. Walk-forward tests on data the strategy never saw. Most "great" strategies die right here. That's the point.
Real results
The gauntlet does its job.
× Killed it
A "copy congress trades" strategy looked great in-sample. Then it fell apart on out-of-sample data in walk-forward. The system failed it. So I dropped it.
✓ Passed
Trend-following passed on SPY, QQQ, GLD and EFA. ~10-year backtest, Sharpe roughly 0.5 to 1.0, and it holds out-of-sample.
Same gauntlet killed one and confirmed the other. That's the system working, not me hoping.
Running right now
It trades the validated edge forward, 24/7.
Only strategies that pass the gauntlet go live in paper. The bot rebalances the validated allocation every day after the close, on its own, and records its equity every tick. That growing forward track record is the one thing a backtest can never give you.
Hosting
Live 24/7
Runs on its own server, daily after close.
State
Durable
Survives restarts and redeploys. Nothing lost.
Money
Paper, for now
Simulated fills. No real capital at risk yet.
Before it ever touches real money
Two walls between idea and execution.
🛡 Risk gate
Every order passes pre-trade checks first: max position size, no leverage, allowed symbols only, cooldowns. A bad order gets blocked, not filled.
⚒ Approval ledger
Orders work like git: stage, commit, push. Nothing executes until an explicit push, with a full reviewable history. The human stays in control.
This is the difference between a demo and something you'd trust with real capital.
The cockpit
One dark command centre.
Markets
BTC/USD+1.4%
SPY+0.3%
QQQ-0.2%
GLD+0.6%
Equity curve // paper bot
Trend allocationSPY · QQQ · GLD · EFA
Backtest lab
Sharpe0.84
Max drawdown-14.2%
Agents
Trend botonline
Edge factoryonline
Gauntletonline
Live multi-asset markets, a backtest lab on the real engine, the gauntlet, the live bot and its track record. All in one place.
Built properly
Not a weekend hack.
Tested
128 tests
Automated, green. The math is checked, not assumed.
Coverage
Multi-asset
Crypto, stocks, FX and macro on one engine.
Status
Live + gated
Deployed, durable, and walled off from real money.
Built to the same standard I'd ship to a client.
Where it's going
The path to real money.
⚛ AI quant brain
Proposes new strategies into the gauntlet, reads the macro regime, writes the daily brief.
⚡ Live execution
Real broker adapters behind hard risk caps and a kill switch. Only once the paper record earns it.
◕ Options income
Wheel and income strategies, added once the core is proven and trusted.
Built with Innoval OS
This is the kind of system I build.
The engine, the validation, the risk rails, the live bot. All of it built solo, fast, to production standard.
Following the build?
I'm sharing the whole thing as it goes from paper to live. The wins, the strategies the gauntlet kills, all of it.
innovalos.com
Educational walkthrough of a personal build. Paper trading only. Not financial advice and not an offer. Past or simulated performance does not predict future results.