Platform Overview

What AIVEX Does

AIVEX is an end-to-end AI research platform that transforms raw market data into governed, auditable signal outputs — without the black box.

How It Works

01

Ingest

Parallel data pipelines pull news, OHLCV data, and fundamental metrics from verified sources on configurable schedules.

02

Analyze

Each atomic signal module applies domain-specific models — NLP for news, technical indicators for charts, ratio analysis for metrics.

03

Compose

The Signal Engine merges outputs with confidence weighting. Conflicting signals are resolved using priority rules and cooldown logic.

04

Govern & Emit

The Governor applies risk thresholds and kill switches. Approved signals are exposed via the Eye REST API with structured JSON.

Key Differentiators

Sub-second latency

In-process signal composition without message queues for the critical path.

Audit trails

Every signal includes a traceable computation path. Full JSON crash records for every exception.

Confidence scoring

Every output includes a 0–1 confidence value computed from source quality and agreement.

Multi-source consensus

Signals require agreement across independent sources before reaching the composition layer.

24/7 supervision

Watchdog process with exponential-backoff restart, crash-loop detection, and heartbeat monitoring.

Research compliance

Built-in disclaimer enforcement. All outputs are research artifacts, never trading recommendations.

Example Output

Real signal record from the Eye REST API (/eye/signals). Fields map directly to the final_signals database schema.

{
  "signal_id":    "sig_4a7f2b3c9d1e",
  "symbol":       "AAPL",
  "direction":    "BUY",
  "confidence":   0.74,
  "strength":     "moderate",
  "sources":      ["news", "chart"],
  "reasons": [
    "Positive sentiment across 3 recent articles (urgency: 0.81)",
    "RSI divergence with price above 20-day moving average"
  ],
  "atomic_count": 2,
  "has_conflict": false,
  "trace_id":     "tr_9e1a4b7f2c3d8e0a",
  "created_at":   "2026-03-30T09:14:22.384Z"
}

trace_id links every signal back to its exact pipeline run. sources lists which atomic modules contributed. confidence is always in [0, 1].

Roadmap

NowNextLater
News, Chart, Metrics modulesAlternative data module (social, satellite)LLM-based signal narrative generation
Signal Engine compositionVector store memory for signal contextBacktesting harness v2
Governor gatingReal-time WebSocket stream APICustom signal module SDK
Eye REST APIMulti-symbol portfolio viewsInstitutional data feed connectors
Watchdog 24/7 runtime