finkurs real-time data terminal displaying predictive portfolio signals

System status: operational

Predictive portfolio logic, deployed in under 60 seconds

finkurs runs latent vector analysis on market data streams and converts the output into position sizing and risk parameters. Connect an exchange API, set your constraints, and let the model execute against deterministic rules — no manual chart reading required.

Initialize Portfolio View Methodology

Median inference latency 42 ms

Data points processed / min 1.8M

Active model instances 3,214

Avg. setup time 54 s

Data Engine

How the prediction pipeline is structured

The engine ingests order-book, volume, and macro data in parallel, then compresses each stream into a latent vector before the stochastic model produces a recommendation. Every stage is logged for audit.

Ingestion layer Streams tick-level price data, order-book depth, and macro indicators from connected exchanges via websocket.

Latent vector encoding Reduces high-dimensional market state into a compact representation the model can reason over efficiently.

Stochastic modeling layer Generates a probability-weighted distribution of likely price paths rather than a single point forecast.

Decision output Translates the distribution into position size, entry range, and stop parameters based on your risk profile.

Multi-exchange normalization Price and volume data from different venues are reconciled into a single internal schema before modeling.
Continuous retraining Model weights are refreshed on a rolling window so the system adapts to regime shifts without manual intervention.
Deterministic execution rules Once a signal crosses your defined threshold, order parameters are fixed and logged before submission.

42 ms

Median inference

6 s

Retraining cycle

99.2%

Feed uptime, 30d

Setup

From API key to deployed model in three steps

The setup removes the manual overhead of connecting exchanges, sizing positions by hand, and re-checking risk limits after every trade. Most accounts complete deployment in under a minute.

STEP 01

API integration

Paste a read-only or trade-scoped API key from your exchange. Keys are encrypted at rest and never leave your account context.

STEP 02

Risk parameter definition

Set maximum drawdown, position size limits, and volatility tolerance. These constraints override any model output at execution time.

STEP 03

Deployment

Confirm the configuration and the model instance goes live. Execution begins on the next qualifying signal from the data engine.

Connect API → Define limits → Model deployed → Signals execute

Decision Layer

The optimization interface, in raw numbers

Every recommendation is shown with its underlying risk-to-reward calculation, not just a directional call. You can inspect, adjust, or override any signal before it reaches execution.

Risk / reward matrix — sample assets Live feed

Asset	Signal	Confidence	Risk score	Expected range
ETH/EUR	Increase	0.78	Medium	+2.1% / −1.4%
DAX Futures	Hold	0.54	Low	+0.6% / −0.5%
BTC/USD	Reduce	0.71	High	+1.2% / −3.0%
EUR/USD	Hold	0.61	Low	+0.4% / −0.3%

Real-time signal feed

02:14:09Volatility threshold crossed — DAX Futures
02:11:47Latent vector shift detected — BTC/USD
02:08:22Risk score recalculated — ETH/EUR
02:05:03Model retrained on 6s window — global

Optimization toggles

Max drawdown cap

Auto-rebalance on regime shift

Slippage guard

Manual override on signals

Methodology

What runs under the recommendation layer

We publish the mechanics of the model rather than asking for trust on faith. The following applies to every account, regardless of asset class or exchange.

Model architecture

The core is a sequence model trained on tick-level and order-book data, paired with a stochastic layer that outputs a probability distribution instead of a single price target.

Latent vector encoding per asset class
Rolling-window retraining, no static snapshots
Separate models for high and low liquidity instruments

Backtesting framework

Every model version is validated against walk-forward backtests using out-of-sample data segmented by market regime, not a single historical period.

Regime-segmented validation sets
Transaction cost and spread simulation included
Version history retained for audit

Slippage protection logic

Orders are checked against live spread and depth before submission. If expected slippage exceeds your tolerance, the order is held and flagged rather than executed.

Pre-trade spread and depth check
Configurable slippage tolerance per instrument
Held orders logged with reason code

Deploy your first model instance

Connect an API key, set your risk parameters, and the model goes live on the next deployment cycle. No manual configuration files, no support tickets required to start.

Deploy AI Instance

GDPR-aligned data handling TLS 1.3 encrypted transport Read-only API scopes by default

API key receivedt + 0s

Risk parameters confirmedt + 18s

Model instance initializedt + 41s

Signal execution activet + 54s