nosyntax

An AI research lab

The quant research process, run by agents.

nosyntax is an AI research lab. Our agents survey the literature, distil it into trade signals and test them on data held out from their design. Those that hold up are reviewed on their evidence, then combined into a single live strategy, weighted to keep correlation low and risk-adjusted returns high.

Papers in, strategies out.

  1. Agent

    Read the literature

    Papers, preprints and market-structure notes are read one at a time. Every claim is pulled out with its evidence and filed on a map of ideas, by mechanism and by the data it needs.

  2. Agent

    Generate signals

    The most promising open ideas become testable signals: what to observe, which way to trade and over what horizon.

  3. Agent

    Backtest

    Each signal is tested on historical data it was not designed on, and every result is written to a research ledger, including the failures.

  4. Agent

    Iterate

    Each round is informed by the last, and the ledger counts every variant tried, so a signal is judged on the whole search, not its best run. Ideas that fail are set aside for good.

  5. Researcher

    Review

    The strongest signals are reviewed with their full trail: the source paper, every round and the numbers.

  6. Agent

    Combine

    Approved signals are checked against each other. Ones that are uncorrelated are combined into a single strategy, so no one idea carries the book.

  7. Agent

    Deploy

    The agent deploys the strategy to its own exchange account. Positions, fills and equity are monitored live from the moment it starts trading.