Reference Architectures

Complete deployment patterns, not just a diagram.

Each architecture shows the before state, where Sentiqra fits, data flow, decision ownership, failure behavior, and economics - so you can adapt it to your stack.

High-volume synchronous safety

Multilingual Content Moderation

Score user content before it goes live; route by severity and policy area.

View architecture

SIEM/SOAR integration

SOC Text Normalization

Normalize reported phishing and threat text into structured signals before SIEM ingest.

View architecture

Conditional multi-domain routing

Marketplace Abuse & Scam

Combine Trust & Safety with fraud signals to screen listings and messages.

View architecture

Routing + emotion + security escalation

Customer Support Routing

Route tickets by intent, urgency, and risk - including security escalation - before an agent reads them.

View architecture

Event-driven analyst pipeline

Crisis Monitoring

Triage 21-language streams with the Crisis Intelligence analyzer - 650 indicators into a prioritized analyst workbench.

View architecture

Deterministic tool layer around LLM

AI Agent Guardrails

Give LLM agents fixed detect_* tools via the MCP server - policy decisions on replayable signals, not prompt prose.

View architecture

News/feed event extraction

Market Intelligence

Tag filings, news, and wires at ingest - 200 indicators across 8 languages into a consistent event taxonomy.

View architecture

These are illustrative reference architectures, not customer case studies. Each shows a recommended pattern you can adapt.