mohamadomar-ai/kairos-engine — explained in plain English
Analysis updated 2026-05-18
Test whether an FX or metals trading strategy survives real broker costs before risking money.
Classify market conditions into regimes like breakout, trend, or chop using a Hidden Markov Model.
Compare forecasts from multiple time series models on the same historical data.
Generate per-regime performance reports to see where a strategy wins or loses.
| mohamadomar-ai/kairos-engine | 0-bingwu-0/live-interpreter | 010zx00x1/faresnipe | |
|---|---|---|---|
| Stars | 2 | 2 | 2 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 4/5 | 2/5 | 2/5 |
| Audience | researcher | general | general |
Figures from each repo's GitHub metadata at analysis time.
Setup script clones several external forecasting model repositories and pretrains a regime classifier, which takes time before the first backtest runs.
Kairos Engine is a research tool for testing trading strategies in currency and precious metals markets, built by Mohamad Omar of USTechai. Rather than trying to sell a winning strategy, its whole purpose is to rigorously check whether a trading idea actually works once realistic costs are included, and to reject it honestly when it does not. The pipeline takes raw, tick by tick price data and runs it through several stages. First, a statistical model called a Hidden Markov Model sorts the market into one of four regimes, such as breakout or trending periods. Then an ensemble of four separate time series forecasting models tries to predict price direction. Candidate strategies are then tested using a walk forward method, meaning the model is only ever tested on data it has not seen before, and every trade is priced using real measured broker costs like spread, slippage, commission, and swap rather than rough estimates. A strategy only counts as validated if it remains profitable after all of that, over enough trades to be statistically meaningful. The README is unusually transparent about failure. It documents a forecast ensemble that showed no real predictive edge, and several strategy variants that lost money once true costs were applied. Only one variant, tested on five minute bars over 221 trades across a full year, passed every check, and even that variant loses money in one of its four market regimes, which the author explicitly says should not be traded. The project includes a setup script and several Python scripts to reproduce the backtests, pretrain the regime classifier, and generate cost sensitivity reports. It is written in Python and depends on the hmmlearn library along with several external time series forecasting model repositories that the setup script clones automatically. This is research and educational software, not investment advice, and no real money has been risked in producing any of its results. It is aimed at people with some background in trading and data analysis who want to test strategy ideas rigorously rather than take them on faith. It is released under the permissive MIT license.
A rigorous backtesting pipeline that tests FX and metals trading strategies against real broker costs and rejects the ones that do not actually make money.
Mainly Python. The stack also includes Python, hmmlearn, Dukascopy API.
Released under the MIT License, which allows free use, modification, and commercial use as long as the original copyright notice is kept.
Setup difficulty is rated moderate, with roughly 1h+ to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.