Enigma

A foundational AI for markets.

Built from first principles. The AI is the quant. AlphaZero for markets.

What Enigma is

The AI is the quant.

Today, quants build the machines that make money. Humans design strategies, write the backtests, read the filings, manage the risk. AI is a copilot.

Enigma replaces the quant. Every step a professional researcher performs - data, hypothesis, code, backtest, deploy, monitor, improve - Enigma performs, end to end. No human in any of those loops.

Then it recursively improves itself. Rewrites its own prompts. Designs its own tools. Invents its own research directions. AlphaZero for markets.

Foundational, not fine-tuned
Enigma is trained from first principles on market data the internet does not contain. Not a wrapper around a frontier model - a dedicated foundation model for financial reasoning.
P&L is the reward
Every stage is evaluated against live P&L under execution constraints. Backtest goodness counts for nothing. Out-of-sample Sharpe counts for everything.
Recursive self-improvement
Enigma rewrites its own prompts, designs its own tools, and invents its own research directions. AlphaGo played Go. AlphaZero learned to play any game. Enigma learns to learn to trade.
Ungameable by construction
Hidden out-of-sample windows. Cascading gates. Net-exposure constraints that prevent the directional bets models default to. If it passes here, it has learned something real.
The loop

Data → hypothesis → code → backtest → deploy → monitor → improve.

01
Data
Ingests market data, filings, order-book state, macro, news, and its own prior trades. Everything priceable is a feature.
02
Hypothesis
Forms testable hypotheses about structure in the data. Not signal discovery by grep - reasoning about why an edge could exist.
03
Code
Writes the backtest, the feature pipeline, the risk overlay, the execution logic. Production-grade software, end to end.
04
Backtest
Runs the hypothesis against history with realistic slippage, borrow costs, capacity constraints, and adversarial counterparties.
05
Deploy
Ships survivors to live capital. Position sizing, correlation management, and kill-switches baked in.
06
Monitor
Watches for regime change, signal decay, and execution degradation. Surfaces its own failure modes before the P&L does.
07
Improve
Feeds live outcomes back into the model. Rewrites its own prompts. Designs its own tools. Invents its own research directions.
Training signal

What we train on that the internet does not contain.

The raw building blocks of a quant exist inside a machine for the first time in history. What's missing is the specialised training.

Eight years of live-executed trades, action-labelled and tagged with P&L outcomes. RL environments that model real order books, slippage, and adversarial counterparties. Expert decisions from professional researchers, annotated with reasoning. The body of knowledge quants spend a career absorbing - captured, structured, and fed to the model.

LLMs can code because the internet is full of code. They can reason about math because the internet is full of math. They can't trade because the internet has no alpha. Enigma trains on the data it does.

Get in touch

Partnering on Enigma or deploying capital against its output?

Contact the teamSee Alpha-Eval