Home › Architecture

Architecture

Infrastructure built for financial-grade precision

Four layers of purpose-engineered infrastructure — from raw market data ingestion to the APIs your systems already speak. Every layer observable, auditable, and deployable on your terms.

Platform layers

Four layers, zero gaps

Layer 04 — Integration

Integration Mesh

Your stack
REST & WebSocketLow-latency event streaming
FIX ProtocolNative capital markets connectivity
Pre-built connectorsMurex, Calypso, Bloomberg, Salesforce
Layer 03 — API

Intelligence API

Unified surface
Signal endpointsRisk, fraud, alpha, suitability
Webhooks & SSEPush-based alert delivery
GraphQL interfaceFlexible query for audit & lineage
Layer 02 — Inference

Inference Engine

Sub-10ms
Purpose-trained modelsFinance-domain, not fine-tuned LLMs
Real-time & batchStreaming inference and scheduled jobs
Model versioningImmutable model registry with rollback
Layer 01 — Data

Ingestion & Normalization

Foundation
Market data feedsL1/L2 tick data, order book, reference data
Document pipelinePDF, XBRL, iXBRL, imaging with OCR
Alternative dataSatellite, NLP, supply chain, web signals
01 / 04

Ingestion & Normalization


Data Layer
50+
Data source types
<5ms
Ingestion latency

Every insight begins with data that arrives correctly, completely, and on time. Our ingestion layer handles the full breadth of financial data formats — from microsecond tick streams to hundred-page regulatory filings — normalizing each source into a unified schema before it ever touches a model.

  • Level 1 and Level 2 market data via direct exchange connectivity and consolidated tape across North American and European markets
  • Document ingestion pipeline supporting PDF, XBRL, iXBRL, Word, and scanned image formats with multi-engine OCR fallback
  • Alternative data connectors for satellite imagery, earnings transcripts, shipping manifests, web traffic, and patent filings
  • Reference data management: instrument master, entity hierarchies, ISIN/CUSIP/SEDOL resolution, and beneficial ownership chains
  • Schema validation, deduplication, and anomaly detection on every inbound record before downstream processing
02 / 04

Inference Engine


Core Intelligence
<10ms
P99 inference latency
99.99%
Uptime SLA

Our inference engine is not a wrapper around a general-purpose model. It is a purpose-built serving layer for financial AI — optimized for microsecond decision paths, stateful streaming inference, and the kind of explainability that regulators actually require.

  • Domain-specific model architectures: graph neural networks for entity resolution, transformer variants for document understanding, and gradient boosted ensembles for real-time scoring
  • Streaming inference over stateful windows — maintaining position-level context across millions of instruments simultaneously
  • Immutable model registry with version pinning, canary rollout, and instant rollback — every production inference is traceable to a specific model version
  • SHAP-based feature attribution attached to every output — explainability is not an add-on, it is part of the response schema
  • Shadow mode evaluation: run new model versions in parallel against production traffic without affecting live outputs
03 / 04

Intelligence API


Unified Surface
6+
API protocols
100%
Audit coverage

A single, consistent API surface across every intelligence capability — whether you are pulling a risk score, streaming surveillance alerts, or querying document extractions. Designed to feel native in the workflows your engineering team already operates.

  • REST endpoints for synchronous request-response patterns: document extraction, entity scoring, portfolio analysis, and signal retrieval
  • WebSocket and Server-Sent Events channels for streaming alert delivery with sub-10ms push latency
  • GraphQL interface for flexible audit log queries, feature lineage traversal, and cross-capability joins
  • FIX protocol adapter for capital markets integrations — native messaging without translation layers
  • Every API call logged with caller identity, model version, input hash, and response — full audit trail by default
04 / 04

Integration & Deployment


Your Environment
3+
Deployment models
40+
Pre-built connectors

Synthetic Reality deploys where your data lives. For institutions with strict data residency requirements, we operate within your cloud tenancy or on-premises environment. For those prioritizing speed-to-value, our managed SaaS tier is operational within days.

  • Managed cloud (AWS, Azure, GCP) with dedicated single-tenant VPCs — your data never co-mingles with another institution
  • VPC-peered private deployment for institutions requiring data residency in Canada, the EU, or specific jurisdictions
  • On-premises deployment via Kubernetes Helm charts with air-gapped model delivery for maximum data sovereignty
  • Pre-built connectors for Murex, Calypso, OpenGamma, Bloomberg Terminal, Refinitiv Eikon, Salesforce FSC, and Microsoft Dynamics
  • Infrastructure-as-code templates (Terraform, Pulumi) and CI/CD pipelines for engineering teams that operate their own deployment toolchain

Technical specifications

Numbers that define the platform

Inference P99 latency
<10ms
Real-time scoring across risk, fraud, and alpha signal endpoints
Ingestion throughput
1M+ msg/s
Sustained tick and event ingestion across all connected sources
Platform uptime SLA
99.99%
Measured across inference, API, and data ingestion layers
Model versions retained
∞
Full immutable model registry — every production version preserved
Deployment options
3+
Managed SaaS, VPC-peered cloud, and fully on-premises Kubernetes
Pre-built connectors
40+
Risk systems, CRMs, trading platforms, and regulatory reporting tools
Data source types
50+
Market feeds, document formats, alternative data, and reference data
Audit log retention
7yr
Tamper-evident, regulator-accessible audit trail by default

Get started

Ready to deploy AI that
understands financial reality?

Join institutions using Synthetic Reality to make faster, smarter, and more defensible decisions at scale.

Request a demo Talk to our team