An AI quant laboratory,
open to everyone.
Combine proven signals into strategy recipes. Let the engine search thousands of variations to improve them. Keep only what survives the evidence. AI researches. You decide. Algorithms execute.Institutional-grade machinery. No private jet required.
- Status
- Proof of concept complete
- Engine
- One core, three modes
- Access
- Private, by application
Most backtests are fiction.
A backtest is the only evidence available before money is at risk. Usually it is wrong, in ways nobody can check.
- F-1Lost in translation
A patchwork of systems, written by different people in different languages. Every hand-off loses something, until the live system can no longer faithfully replicate the backtest.
One deterministic core for every mode. See Section 03
- F-2Over-fitting
Test enough variations and one will look brilliant by pure chance.
Every trial counted. Results judged against the number of attempts. See Section 04
- F-3Hidden bias
The future leaks into the past. The periods where a strategy would have failed quietly go missing.
The market replayed tick by tick, exactly as it happened. See Section 03
- F-4No audit trail
Months later, nobody can reproduce the result. The code, the data or the settings changed.
Every run recorded and reproducible, byte for byte. See Section 07
For an auditor, an unreproducible backtest is not evidence. For an individual, it is a dangerous trap.
- Regulators & auditors
- Compliance by design. Every decision, every trade, every byte on the record and replayable on demand. The audit trail regulators have always wanted, built in from day one.
- Researchers
- A gold mine. Real-market experiments anyone can reproduce, challenge and publish: a laboratory for finance that finally meets the standards of science.
- Individuals
- Protection by design. Nothing touches real money until it has been proven, and every risk is explained before it is taken.
The backtest is the trade.
Portfolium runs a single deterministic core in all three modes: backtest, paper and live. Only the market feed and the exchange connection change.
Identical inputs produce byte-identical outputs. A validated backtest is not an optimistic estimate. It is a rehearsal.
Backtest
- Feed
- Recorded ticks
- Exchange
- Simulated
Paper
- Feed
- Live market
- Exchange
- Simulated
Live
- Feed
- Live market
- Exchange
- Real exchange
Deterministic core · one implementation
signals · recipes · sizing · exits · interlocks · ledger
Output
7f3a…e91c
Output
7f3a…e91c
Output
7f3a…e91c
Fig. 2One engine, three modes. Only the feed and the exchange connection change. The outputs are compared by hash.
Signals are ingredients. Strategies are recipes.
AI proposes.
Evidence decides.
In Portfolium, a strategy is not hand-built code. It is a recipe: market signals, sizing rules, defences and exits, combined like ingredients. Write one yourself, import one from TradingView, or let the research engine propose new ones.
A self-improving search then runs thousands of experiments across every parameter, learning from each round where to look next. AI does the searching. It never does the trading. Every result must pass the same statistical tests the largest quant funds use to separate skill from luck.
- Signal
- Reversal tendency · multi-scale
- Regime
- Volatility filter
- Sizing
- Finite ladder · exposure-budgeted
- Exit
- Adaptive trailing stop
Fig. 3A recipe card. Generic ingredients shown. Each is a tested, reusable component of the engine.
Fig. 4Self-improving search (schematic). Each generation samples around the best of the last. Isolated peaks are struck out: they are usually luck.
Edge or luck? Eight gates decide.
Every strategy, whether you wrote it or the AI found it, must clear the same eight gates. The bar is set before the results come in, using the tests institutional researchers rely on to expose luck and over-fitting.
- 01Component trust
- 02Baseline survey
- 03Optimisation
- 04Statistical validation
- 05Out-of-sample robustness
- 06Reproducibility
- 07Paper trading
- 08Live trading
Fig. 5Gate sequence. Any gate fails: back to research. Never forward.
- Armed
Isolated sub-accounts
Every position is ring-fenced in its own sub-account.
- Armed
Daily loss limit
New entries stop once the day's limit is reached.
- Armed
Drawdown tiers
Size reduces, then trading halts, as drawdown deepens.
- Armed
Performance kill switch
Live is compared with its validated reference. Divergence halts the strategy.
- Armed
Emergency stop
Held by the user. One message halts trading at once.
Fig. 6Interlock panel. Attached by the core to every account. There is no switch to turn them off.
Risk is infrastructure, not strategy.
Every account runs inside controls the engine cannot switch off. No strategy is exempt, however good its backtest.
Leverage is treated as a precision instrument, not an amplifier.
From first lesson to fully automated.
Portfolium is a laboratory that teaches. A newcomer starts with what a portfolio is, then risk, then trading, then leverage. A professional goes straight to the recipes, the search and the statistical reports. It is the same laboratory and the same engine at every level.
01
Learn
Portfolio, risk and trading, from first principles. An AI tutor answers every “but why?”.
02
Practise
Paper trading on live prices, on the engine that trades real money.
03
Experiment
Build recipes, import TradingView strategies, run the search.
04
Automate
Validated strategies, running around the clock, on your own accounts.
05
Carry
Updates, explanations and the emergency stop, in Telegram and WhatsApp.
Fig. 7The operator's path. Each stage uses the same engine as the next, so practice is realistic.
Algorithms trade,
AI researches and teaches.
Every trade comes from a deterministic, validated algorithm that can be audited to the byte. The AI refines strategies, explains, tutors and interprets results, so the person in charge makes better decisions. It never places a trade.
Proof of concept: complete.
The hard part is built. The engine trades live, with real money, on exchange. Its backtests have been shown to match its live decisions.
Inspection record · Model AQE-1 · Proof of concept
- 01Passed
Determinism
Identical inputs reproduce byte-identical outputs
Evidence: Repeated runs, hash comparison
- 02Passed
Backtest = paper
Replaying a paper session reproduces its decisions and fills
Evidence: Trade-by-trade comparison
- 03Passed
Simulated = real exchange
Margin, fees, funding and liquidation match the exchange
Evidence: Exchange-recorded reference data
- 04Passed
Paper = live
Parallel sessions on the same market make the same decisions
Evidence: Reconciliation report
- 05Passed
Live trading
Real money, minimum size, fully reconciled
Evidence: Exchange statements vs ledger
- 06Passed
Statistical soundness
Validation passed on fresh data, thresholds declared in advance
Evidence: Validation report
A reproducible evidence pack is available to qualified parties under NDA.Request it →
Request early access.
Access is staged and reviewed by hand. Tell us who you are.
- Investors
- The evidence pack is available under NDA.
- Direct
- contact@portfolium.ai
Be among the first to try the platform.
Not an investor or partner? Join the waiting list. Places open in stages, and we will write to you as they do.