AlgoVesta v38 — Full Technical Architecture

14 Proprietary Engines. AI Agents.
Institutional-Grade Intelligence.

500+ coin, 14 analiz motoru, 13 yapay zeka ajanı — hepsi aynı anda çalışır. Her işlem kararı 17 veri kaynağından doğrulanır, 0.3 saniyede borsaya iletilir.

14 Analiz Motoru
8 Borsa CVD
537 Coin
24/7 AI Ajan
DECISION CENTER

Multiplicative Gate Scoring

Instead of additive scoring, AlgoVesta uses a multiplicative gate system. CVD, regime, volume, and institutional pressure gates are multiplied to produce the final score. If any gate falls below 0.3, the signal is automatically rejected — no trades on conflicting signals.

Gate Multipliers (final = base × gate1 × gate2 × ...)

Min Gate: 0.3
Critical Gate Multipliers (0.0 – 1.0)
CVD Gate
0.92
Regime Gate
0.88
Volume Gate
0.95
Institutional Gate
0.85
ML Layer Predictions
MLP Win Prob
0.72
HMM Viterbi Conf.
0.89
Bayesian GP EI
0.34

Multiplicative Gate Formula: final_score = base_score × cvd_gate × regime_gate × volume_gate × institutional_gate. If any gate drops below 0.3, the signal is automatically rejected. MLP Neural Network (28→64→32→1) predicts trade success probability, HMM Viterbi decodes regime transitions with O(T×N²).

TECHNICAL ARCHITECTURE

14 Proprietary Analysis Engines

Each engine specializes in a different market dimension and produces independent signals. All outputs flow into the Neural Scoring Matrix to form the final decision. No single engine's false signal can become a trade without validation from the others.

01

CVD Trend Engine

Cumulative Volume Delta

Collects volume-weighted CVD data from 8 exchanges (Binance, Bybit, OKX, Bitget, dYdX, Hyperliquid, GMX, Vertex). Multi-window analysis (5m, 15m, 1h) detects confluence.

Volume-Weighted 3-Window CVD Slope Normalize
02

Momentum Engine

Velocity & Acceleration

Measures price velocity (%/min) and acceleration (dVelocity/dt). Detects trend exhaustion points to prevent late entries.

Velocity Acceleration Exhaustion
03

Whale Tracker

Institutional vs. Retail Flow

Classifies each trade by symbol-specific thresholds (BTC $200K, ETH $150K, SOL $100K). When institutional money buys while retail sells, a strong BUY signal is generated.

Symbol Thresholds Whale Flow Real-Time
04

VPIN Toxicity

Probability of Informed Trading

Volume-Synchronized Probability of Informed Trading. NumPy-vectorized analysis over 50 volume buckets detects when institutional players make the market toxic. Vetoes when VPIN ≥ 0.75.

50 Volume Buckets NumPy 0.75 Veto
05

Funding Dynamics

Funding Rate Analysis

Evaluates funding rates in two tiers: +0.03% signals excessive long crowding (top risk), -0.03% signals excessive short crowding (bottom opportunity). Contrarian bonus for aligned signals.

Contrarian Signal Cost/TP Veto
06

Liquidation Gravity

Liquidation Gravity Map

Maps liquidation clusters via Coinglass data. Identifies where price is gravitationally pulled toward dense liquidation zones. Detects squeeze setups before they trigger.

Coinglass API Squeeze Detection Gravity Score
07

Long/Short Ratio

Long/Short Ratio Analysis

Tracks long/short ratio from Coinglass to detect market positioning extremes. Extreme ratios indicate crowded trades ripe for reversal.

Coinglass API Positioning Contrarian
08

Exchange Netflow

Exchange Inflow/Outflow

Monitors net flow of assets into and out of exchanges via Coinglass. Large inflows signal potential sell pressure; large outflows signal accumulation.

Coinglass API Inflow/Outflow Accumulation
09

Hawkes Process

Event Clustering Detection

Models self-exciting event processes to detect trade clustering. When trades beget more trades in rapid succession, it signals momentum ignition or institutional activity.

Self-Exciting Clustering Momentum
10

OI Velocity

Open Interest Change Analysis

Analyzes the rate of open interest change combined with directional signals. OI increase + BUY signals new money inflow (+10% bonus). OI decrease + BUY carries squeeze risk (0.80x penalty).

OI Change Rate Money Inflow Direction Alignment
11

Signal Decay

Age-Based Weight Reduction

Applies time-based decay to signal weights. Older signals lose influence exponentially, ensuring the system always prioritizes the freshest market data.

Exponential Decay Freshness Priority
12

Order Book Imbalance

Order Book Wall Detection

Scans 20 levels of the order book. Orders exceeding 5x the average level are flagged as walls. BID walls block SHORT; ASK walls block LONG entries.

20 Levels 5x Threshold Direction Block
13

Forced Transaction Detector

Forced Transaction Detection

Combines 4 signal sources to detect forced transactions: liquidation cascades, margin calls, and stop-loss hunting patterns. Identifies mechanical price movements before they complete.

4-Signal Fusion Cascade Detection Pre-Completion
14

SL Cluster Map

Stop-Loss Cluster Mapping

Maps stop-loss cluster zones where large groups of traders have placed their stops. Identifies high-probability reversal zones and avoids placing stops in vulnerable clusters.

Cluster Zones Reversal Zones Smart SL
COIN SELECTION SYSTEM

Pre-Scanner Anomaly Radar

An intelligent filtering system that selects real opportunities from a pool of 537 coins. Cuts through noise and focuses resources on the strongest targets.

1

537 Coin Pool

537 coins classified across 11 tiers. Hierarchical scanning from Tier 1 (BTC, ETH) to Tier 11 (low volume).

2

Anomaly Filter

Detects coins moving independently from BTC. Calculates anomaly score using volume velocity + price deviation.

3

Top 10 Targets

The 10 coins with the highest anomaly score are sent to the Neural Scoring Matrix. 5 coins are scanned in parallel simultaneously (ThreadPoolExecutor).

8-Exchange CVD Consensus

Binance (0.30) Bybit (0.20) OKX (0.15) Bitget (0.10) Gate.io (0.08) MEXC (0.07) HTX (0.05) KuCoin (0.05)

Weights: weight_i = volume_i / sum(all_volumes). 3-sigma anomaly detection disables manipulative exchanges.

● AJAN KONSEYİ

13 Uzman. Tek Sistem.

Her işlem kararı bir yönetim kurulu toplantısı gibi çalışır. 13 farklı uzmanlık alanından ajan aynı soruyu masaya getirir. ≥8/13 onay sağlanmazsa işlem açılmaz — bu kural hiç bozulmadı.

0.3 sn
Karar hızı
≥ 8/13
Konsensüs eşiği
4
Uzmanlık departmanı
01 · BAŞ STRATEJİST

Brain

Piyasanın yönünü belirler. Al/Sat/Bekle — son söz onda.

02 · VP · FIRSAT

Hunter

500+ coin arasında hareketi yakalar.

03 · VP · RİSK

Watchdog

Sistem sağlığını 7/24 izler.

05 · VP · AR-GE

Darwin

Ayarları evrimleştirir, iyileştirir.

12 · VP · KONSENSÜS

Diplomat

Ajanları oylamaya sokar.

07 · UZMAN

Oracle

Volatiliteyi tahmin eder.

11 · UZMAN

Architect

Uzun vade trend yapısı.

06 · UZMAN

Sentinel

Aşırılıkta fren yapar.

10 · UZMAN

Medic

Servisleri tamir eder.

04 · UZMAN

Doctor

Hataları teşhis eder.

13 · UZMAN

Mirror

Sistem yansıma doğrulama.

08 · UZMAN

Psychologist

Duygu dengesi gözcüsü.

09 · UZMAN

Accountant

Portföy dengesi.

· UZMAN

TraderTwin

İşlem ikizi simülasyonu.

● MAKRO KORELASYON

Kripto izole değildir.

DXY güçlendiğinde BTC genellikle düşer. AlgoVesta bunu bilir ve pozisyon açmadan önce kontrol eder.

DXY
Dolar Endeksi

Güçlü dolar = risk-off sinyal

Gold
Altın

Güvenli liman talebi tespiti

SPX
S&P 500

Risk iştahı barometresi

VIX
Korku Endeksi

Aşırı volatilite filtresi

Bu 4 makro gösterge eş zamanlı izlenir. Korelasyon eşiği aşılırsa işlem otomatik ertelenir.

● PLATFORM

Sadece bot değil. Tam platform.

📄

Paper Trading

Gerçek para kullanmadan canlı piyasa verisiyle test edin. Strateji kanıtlanmadan canlıya geçmeyin.

💚

Bot Health Score

Botunuzun sağlık durumu 0-100 arasında puanlanır. Düşerse sistem sizi uyarır.

🔍

Signal Transparency Score

Her sinyal için hangi motorların tetiklendiği, hangi ajanların onay verdiği gösterilir.

🏆

Performance Certificate

Aylık performansınız PDF olarak indirilebilir. Paylaşılabilir, doğrulanabilir.

📧

Haftalık AI Email Raporu

Her hafta botunuzun performansı, en iyi/kötü işlemler ve öneriler e-posta ile gelir.

🐋

Whale Alert

Büyük cüzdan hareketleri WebSocket ile anlık izlenir. Kurumsal hareket öncesinde sinyal üretilir.

RISK MANAGEMENT

ATR-Based Risk Management + 8 Safety Layers

ATR-based volatility-adjusted position sizing: position = (risk × balance) / (ATR × multiplier). Automatic downsizing in high volatility, opportunity capture in low volatility. Plus 8 independent safety layers with veto authority to filter signals.

1

Margin Check

No trade is opened without sufficient margin. Minimum margin threshold is configurable (default: $400).

2

Loss Streak Protection

After 3 consecutive losses, a graduated cooldown system activates. 3 losses = 30min pause, 5 losses = 2 hour pause. Automatic cooldown period.

3

Correlation Filter

Only 1 position from each of 6 asset groups (Layer1, DeFi, Layer2, Meme, CEX Token, AI) at a time. No double risk in the same sector.

4

Spread Check

No trade is opened if the bid-ask spread is too high. Prevents slippage risk during low liquidity periods.

5

Volume Confirmation

Current volume must be at least 120% of the 20-candle average. No trades on low volume — signals must be backed by market participation.

6

5-Minute Volatility

If price movement in the last 5 minutes is excessive (sudden pump/dump), the trade is deferred. Prevents late entries.

7

Candle Direction Confirmation

LONG signals require the last candle to be green (close > open), SHORT requires red. Signal must align with current price action.

8

Macro News Calendar

All trades are automatically paused within ±30 minutes of FOMC, NFP, CPI, and PPI announcements. 2026 calendar is built-in.

Silver Vanguard Gate — 7 Additional Filters

Second defense line that runs after the main pipeline

1. CVD Slope positive?
2. Consensus ≥65%?
3. VPIN < 0.75?
4. OI Organic?
5. NLP Sentiment ≥0.35?
6. MTF aligned?
7. No L2 Wall blocking?

Signals that don't pass at least 5 of 7 filters cannot become trades, even with Neural Scoring 72+.

AUTONOMOUS LEARNING

Self-Healing AI + Continuous Retrain

The bot analyzes its own decisions, learns from mistakes, and auto-updates its parameters. 24-hour batch retrain cycle continuously trains BiLSTM and RL models. Shadow mode backtests new models before going live.

Self-Healing Cycle

1

Signal Rejected: Neural Score stayed below 72 or a safety layer vetoed.

2

30-Minute Wait: The rejected signal is recorded and price movement is checked after 30 minutes.

3

Simulation: "If this signal had become a trade, would it have been profitable?" is answered.

4

Calibration: If a missed opportunity is confirmed, the relevant layer weight is adjusted. Layers that incorrectly vetoed profitable signals are penalized.

Adaptive Trade Modes

Micro-Scalp

Tight SL and rapid exits during low ATR periods. Targets small but frequent gains.

Standard

SL = 1.2×ATR, TP = 2.8×ATR. R:R ≥2.0 required. The default and most frequently used mode.

Trend-Ride

Wide SL and trailing TP during strong trend periods. Designed to ride the trend to completion.

Smart Trading Hours (UTC+3)

Green Hours (Active)

10:00-13:00 and 15:00-01:00. High liquidity and volume periods. US and European market overlap hours.

Red Hours (Passive)

01:00-10:00 and 13:00-15:00. Low liquidity, high spread risk. The bot does not open new positions during these hours.

MACHINE LEARNING

4-Layer ML Architecture

Pure-Python implementation — zero external dependencies. Each layer operates independently and is continuously trained via a 24-hour batch retrain cycle. New models are backtested in shadow mode before going live.

L1

MLP Neural Network

Trade Outcome Predictor

3-layer MLP (28→64→32→1) predicts the success probability of each signal. Xavier init, ReLU activation, SGD momentum with backprop. Active after 1000+ trades, shadow mode before that.

28 Feature Input L2 Regularization Shadow Mode
L2

Bayesian GP Optimizer

Dynamic Parameter Optimization

Gaussian Process + RBF kernel automatically discovers optimal parameters (ATR multiplier, SL/TP ratio, minimum score threshold) for each regime. Uses Expected Improvement acquisition function.

RBF Kernel Expected Improvement Regime-Based
L3

Mini LSTM + HMM Viterbi

Regime Transition Predictor

Mini LSTM (4→16→4) predicts regime transitions. HMM Viterbi decoding backtracks hidden Markov states via O(T×N²) dynamic programming. Gaussian emission model with online updates.

Forget/Input/Output Gate Viterbi Backtrack 20-Step Sequence
L4

Batch Backtest & Retrain

Thread-safe Async Pipeline

Automatic 24-hour retrain cycle. Thread-safe daemon thread trains models without blocking the main pipeline. Shadow mode automatically promotes new models that outperform the previous version.

24h Retrain Shadow Mode Thread-safe
END-TO-END PROCESS

How Does a Trade Open?

The bot runs in a continuous loop. Each cycle completes the following 5 steps in sequence.

1

Pre-Scanner: 537 Coin Scan

BTC-decoupled anomaly detection selects the top 10 targets from a pool of 537 coins.

2

14 Engine Parallel Analysis

CVD, Momentum, Whale Tracker, VPIN, Funding, Liquidation Gravity, L/S Ratio, Netflow, Hawkes, OI, Signal Decay, Order Book, Forced TX, SL Cluster — each engine produces independent signals.

3

Multiplicative Gate Scoring + ML Prediction

final = base × cvd_gate × regime_gate × volume_gate. MLP win probability, HMM Viterbi regime confidence, and Bayesian GP parameters are validated.

4

Safety Layers + Silver Vanguard Gate

8 safety checks and 7 additional Silver Vanguard filters. A total of 15 independent veto points.

5

Trade Executes + ML Retrain Cycle

ATR-based position sizing, TWAP/VWAP entry. 24h batch retrain continuously updates MLP/LSTM/GP models.

Telegram Signal Automation

Beyond the AI Bot, a separate automation layer processes Telegram channel signals into live trades. Unlimited channels, AI signal parsing, execution across exchanges.

Telegram Automation Details
v38 — Institutional AI Trading Engine

Start Your 14-Day Free Trial

14 gün boyunca tüm özellikleri ücretsiz deneyin. Kredi kartı gerekmez. İstediğiniz an iptal edin.