SECTION B - AI , quantitative based
Research and developments
AI
AI-Driven Systems. Human-Directed Strategy.

At the core of our development philosophy lies a rigorous approach to research, powered by advanced artificial intelligence and iterative refinement. Every model, tool, and module within our ecosystem is the result of systematic code generation, real-time testing, and deep optimization under realistic execution environments.

All results, findings, and frameworks remain proprietary and are currently restricted for internal application only. While certain components may be offered to third parties in the future, none of the information presented herein constitutes an active commercial product, nor is any portion intended for public solicitation at this stage.

The contents reflect ongoing R&D initiatives and may include experimental features not yet validated in production. No assurance is made regarding current operability or readiness for external deployment.
THE ALPHA CORE
Multiple strategy engines (ML , META)
ALPHA STACK: Strategy HYBRID
The SILTHRA Alpha Stack operates as a modular, hybrid engine for discovering and deploying alpha in intraday equity markets. It combines rule-based strategies with adaptive learning modules, balancing hard-coded execution logic with probabilistic decision systems. Strategies are defined by clear signal-event logic (e.g., RSI band traversals) but are enhanced by meta-models that learn when, where, and how these signals perform best. The hybrid nature refers not just to signals, but to the full cycle: from discovery to deployment.

Memory
The system includes a persistent memory module that logs outcomes, contexts, and parameter interactions for every trade and trial. This memory is not limited to backtests — it learns from live deployments. All data is stored in a structured SQLite schema, supporting fast recall and attribution analysis. Over time, this memory evolves into a database of contextual intelligence: which parameters performed best under which volatility regimes, time-of-day windows, or RSI structures.

Smart Portfolio Management
Position sizing is not fixed. The stack dynamically adjusts trade size using probabilistic estimates of signal quality. It employs reinforcement-based scoring to determine:
• When to reduce risk exposure
• When to scale up with leverage
• When to skip trades entirely due to low context confidence
The system integrates both outcome memory and live signal structure to determine capital allocation. Smart sizing allows conservative capital use in ambiguous regimes, and aggressive posture when historical patterns align with live context.
Models & Learning
A lightweight deep learning model (RSI pattern classifier) works in tandem with gradient-based models to estimate:
• Target hit probability
• Trade duration distribution
• Risk-reward asymmetry per pattern
These models are continuously updated with new trade outcomes. They are not used to generate signals, but to filter and modulate entries produced by the base strategy. The learning component helps the system reject false positives, reduce churn, and improve capital efficiency.
Backtesting
All strategies are evaluated through multi-stage, walk-forward validation with strict separation of training and testing windows. Each stage reduces overfitting by:
• Using different data regimes
• Promoting only top performers
• Re-testing final candidates across new, untouched data
This philosophy assumes markets are non-stationary and adversarial. As such, models and strategies are tested not for performance in the past, but for generalization under stress. No strategy is accepted purely based on in-sample results. Every winner must prove itself out-of-sample.
Live Engine
Live execution is isolated from research layers. The trading engine is deterministic, minimal, and hardened. It operates in shadow or live mode, recording every decision, rejection, and trade result. Entry/exit rules are pre-compiled and version-controlled. The live engine supports:
• Margin allocation with cap
• Latency-sensitive signal timing
• Live pattern validation pre-entry
It does not “learn” during execution but logs everything back to the memory layer for future analysis.

Risk Management
Risk is enforced at every layer:
At strategy level: Minimum hit rate, max drawdown, and minimum trade count gates are enforced.
At sizing level: All positions are capped by max leverage, risk per trade, and volatility-adjusted exposure.
At execution level: Trade rejection logic is used if context deviates from historical safe zones.
The system also tracks tail-risk behavior over time and reduces allocation to strategies that show poor convexity or high variance-to-edge ratios.

Shirvani Group ... html, body { margin: 0; padding: 0; overflow-x: hidden; background-color: black; /* ensures no white */ } .container { max-width: 100%; box-sizing: border-box; }
Legal notice The information provided on this website is for informational purposes only and does not constitute an offer, public offer, invitation or advertisement to attract investments in any form. Shirvani Group LLC operates in the field of proprietary trading, conducting all operations exclusively at its own expense and in full compliance with the legislation of the Russian Federation.: • Federal Law No. 39-FZ "On the Securities Market" (04/22/1996) regulates activities on the securities market and investment activities.• The Civil Code of the Russian Federation establishes the rules of ownership and contractual obligations, including the right to own and carry out transactions with securities.• Federal Law No. 14-FZ "On Limited Liability Companies" (02/08/1998) – regulates the activities of LLC, including the conduct of proprietary trading operations.• The Tax Code of the Russian Federation defines the procedure for taxation of corporate transactions.Additionally, Shirvani Group LLC strictly adheres to the provisions of: • Federal Law No. 38-FZ "On Advertising" (13.03.2006) – excludes the distribution of advertising materials related to attracting investments.• Federal Law No. 115-FZ "On Countering the Legalization (Laundering) of Proceeds from Crime and the Financing of Terrorism" (08/07/2001) – aimed at preventing illegal financial transactions.Shirvani Group LLC does not attract investments from third parties, does not provide financial or investment services, and does not disclose information about current market positions and strategies to third parties.This notice is intended to clarify the nature of the company's activities and prevent any possible misinterpretation of its operations.
Moscow, Golovinskoe highway, 1

Тел. +7 (495)532-54-84

info@shirvanigroup.ru



TAX ID: 7743382169

REG. : 1227700309451

64.99.1

Investments in securities