How we build
Self-tuning engines, tested honestly
Every Helm Quant indicator is built the same way: define a space of possible behaviours, let the data decide which one fits each instrument, then prove it on real history with the drawdowns shown. No magic numbers. No cherry-picking.
Built for myself first
I trade Indian markets every day. These engines exist because I wanted them on my own charts — not as products first. They only go on sale after they survive my own honest backtests, and I keep trading them after they ship.
That is the whole positioning. Premium tools, never budget. Real backtests, not fantasy numbers. Annual plans grandfather their rate forever. And the things a serious trader should be suspicious of — tip groups, “guaranteed returns,” courses on how to trade — are exactly the things Helm Quant will never sell.
How the engines work
01
Self-tuning, every bar
Each engine carries a space of possible behaviours rather than one fixed recipe. On every bar it re-selects the configuration that has been working best on the symbol and timeframe in front of it. That is the adaptation you see live on the chart — no manual tuning, no fragile presets.
02
Learned from data, off-chart
Before an engine ships, it is studied across years of history and hundreds of instruments to learn which setups produce the highest-quality trades for a given kind of market — and hardened so the edge holds on instruments it has never seen. Machine learning does the searching; the engine carries the conclusions.
03
Honest by default
Every engine is proven on real data before it is sold, with drawdowns shown next to returns and execution costs estimated. We publish what a tool does and how we tested it — never the internals that would let the edge be copied.
How we backtest
The standard every engine ships against. ATE is the first to clear it — its full report is published openly, and the other three are measured the same way before launch.
Across many instruments
Tested on real historical data spanning asset classes a serious trader actually watches — Indian indices, US indices, commodities, crypto, and FX — across multiple timeframes. The whole grid is published: the cells that worked and the cells that did not.
Drawdowns beside returns
A return means nothing without the pain it took to earn it. Maximum drawdown sits next to every P&L figure, and the worst-behaving cells stay in the report rather than being quietly dropped.
Costs are not hidden
Headline numbers are gross of brokerage and slippage. Every report ships with a live-drag estimator so you can subtract realistic execution costs for your own broker — and we point out where those costs quietly eat the edge.
Sized like a real account
The simulation risks a fixed percentage of equity per trade and compounds — no fantasy fixed-rupee sizing, no leverage assumptions. When a bar gaps past the stop, the fill is modelled at the open price, worse than the stop level, the way real markets behave.
No small-sample flukes
Configurations with too few trades are excluded from selection, so a three-trade accident never gets promoted to “the strategy.” Thin-sample cells are flagged, not celebrated.
Validated out-of-sample
Each engine is checked on hundreds of instruments it never trained on, across market caps — to see whether the edge survives on unseen data, not just the history it was tuned on.
How we use machine learning
Markets are not all the same, so the best setup for Nifty is not the best setup for gold or for Bitcoin. Rather than guess one recipe and hope it travels, we let the data find the right setup for each instrument.
Off the chart, each engine goes through large machine-learning studies across years of history and hundreds of instruments. The study learns which conditions precede the highest-quality trades for a given kind of market, and which patterns only ever lost money — then distils that into robust rules the engine carries as its defaults. On the chart, the engine keeps adapting bar by bar, choosing its best-performing configuration for whatever you have loaded.
We are deliberately open about the philosophy and silent about the mechanism. The instruments we tested, the way we measure honestly, the fact that the edge is tuned per market — all public. The features, the model, and the numbers behind the tuning stay private, because that is the part that would let the edge be reverse-engineered.
What this is not
No paid signal groups. No tip channels. No “guaranteed returns.” No courses on how to trade. Just tools I use myself, with backtests and drawdowns shown honestly — sold to people who want to run them on their own charts and make their own calls.
The four engines
ATE — Adaptive Trend Engine
Trend-followingRides sustained directional moves and steps out when they turn. The only engine with a published, honest backtest — 10 instruments × 5 timeframes, drawdowns shown next to returns.
Backtested · in development
ACE — Adaptive Consensus Engine
Consensus / confirmationWaits for several independent signals to agree before committing — fewer, higher-conviction trades. Built to run uncorrelated alongside ATE: when one waits, the other often trades.
In development
ASE — Adaptive Strength Engine
Momentum & strengthReads the conviction behind a move and acts at the extremes. The most active engine in the lineup — built for liquid instruments where opportunities come often.
In development
ARE — Adaptive Range Engine
Range & mean-reversionWorks the range when price is not trending — fading the extremes back toward equilibrium. The natural complement to ATE.
In development