Kunara Celvili analyses market microstructure in real time and ranks AI-generated strategies by predictive variance and historical drawdown, so capital can be routed into models with a demonstrated risk-adjusted profile instead of a single discretionary call.
Kunara Celvili was built as infrastructure for traders who already understand systematic execution but lack the time to backtest and monitor dozens of strategy variants in parallel. Instead of issuing isolated signals, the platform ranks an active pool of AI-generated strategies on a rolling basis and exposes the parameters behind each ranking.
The underlying engine is maintained for the requirements of the DACH market, including local trading-hour liquidity patterns and the data latency constraints relevant to execution on European venues.
Order flow, cross-asset correlation, and macro data now update faster than a single analyst can reconcile them into a coherent position. Fatigue and recency bias compound the problem over a full trading session, even for experienced practitioners. Algorithmic execution does not remove uncertainty, but it removes the inconsistency that comes from applying a different threshold to the same setup at 9:00 and at 15:00.
| Dimension | Manual Analysis | Algorithmic Execution |
|---|---|---|
| Reaction time to new data | Seconds to minutes | Sub-second, continuous |
| Exposure to emotional bias | Present, variable by session | Structurally excluded |
| Concurrent data streams processed | Limited by attention span | Parallel, cross-asset |
| Consistency across sessions | Fluctuates with fatigue | Governed by fixed parameters |
The engine does not output a single buy or sell signal. It outputs a ranked distribution of strategies, each carrying its own predictive variance band and drawdown history, which the user reviews before allocating capital.
Predictive variance, not point forecasts, is the primary output surfaced to users. A model that predicts a single future price is making a claim that is difficult to validate; a model that predicts a distribution of plausible outcomes can be checked against realized results over time, which is the basis for how strategies are ranked on this platform.
Copy-trading on Kunara Celvili means allocating a defined share of capital to a specific AI strategy from the ranked pool, with execution handled automatically within the parameters that strategy was tested under.
Browse active AI strategies sorted by risk-adjusted return, trailing drawdown, and predictive variance band.
Open a strategy to view its backtesting window, Sharpe ratio calculation, and the asset classes it trades.
Set a position size as a fixed percentage of your account, with maximum exposure limits applied automatically.
Track live execution against the backtested expectation and reduce or pause allocation if variance exceeds the stated band.
Each strategy is evaluated against historical order-book and price data using a walk-forward methodology, meaning the model is tested on data it was not trained on, in sequential time blocks. This is intended to limit overfitting to a single historical period. Backtest results are labeled with the exact date range and asset universe used, and are kept separate from any live, forward performance tracking shown in the dashboard.
Latency depends on the venue and asset class, and is displayed per strategy as a rolling average. The engine is designed to minimize the gap between model output and order placement, but network and broker-side conditions remain outside the platform's direct control.
The models ingest price and order-book data from connected venues, together with macroeconomic release data and volatility indices. Each strategy's documentation lists the specific data universe it was trained and backtested against.
Integration is handled through a connected brokerage API. Users retain custody of their funds with their own broker; Kunara Celvili sends allocation and execution instructions but does not hold client capital directly.
Yes. Multiple strategies can run concurrently, subject to the correlation check that limits combined exposure to similar risk factors across the active pool.
If live variance exceeds the strategy's stated predictive band beyond a defined threshold, allocation to that strategy is automatically paused pending review, and the event is logged for the user.
Request a protocol overview to see the backtesting documentation, parameter definitions, and current strategy pool before connecting a live account.
Read the Methodology