RateGuard System Architecture V2

Combining deterministic software calculation with agentic AI orchestration on Google Cloud.

End-to-End System Topology & Integration Flow

User Browser / Client → Next.js 14 Frontend

↓ HTTP REST API

FastAPI Backend (Cloud Run API)

↓ Pub/Sub Async Queue / Native Invocation

Private Worker Runtime (Cloud Run Worker)

↓ Agentic Assurance Supervisor

Google GenAI SDK Structured-Decision Supervisor ↔ Gemini 3.7 Flash (Google Vertex AI)

↕ Deterministic Boundaries

Python Deterministic Engines (AST Diff, Oracle Math, Test Generator, Reconciliation)

↓ Cloud Persistence

Firestore (RunState) | BigQuery (50K Portfolio SQL) | GCS (Artifacts)

↓ Release Decision

PASS / REVIEW_REQUIRED / BLOCK_DEPLOYMENT

Strict Deterministic Boundary Guarantee

Gemini 3.7 Flash, invoked through the Google GenAI SDK, plans investigation probes, reasons over AST diffs, proposes remediation patches, and synthesizes executive summaries. All monetary arithmetic, rate table lookups, pricing graph evaluations, test executions, BigQuery SQL aggregations, and financial exposure calculations are executed exclusively by deterministic Python code using Python Decimal.

Google GenAI SDK + Gemini 3.7 Flash

Agentic Supervisor

Orchestrates adaptive assurance missions, interprets findings, plans probes, synthesizes root causes, and proposes remediation patches.

BigQuery (50K Portfolio)

Portfolio Analytics

Stores 50,000 synthetic carrier policies and executes high-performance SQL exposure queries measuring financial blast radius.

Firestore (Native)

Workflow State & Audit

Persists real-time assurance mission state, event logs, and audit evidence records in Native mode database.

Cloud Storage (GCS)

Artifact Management

Stores pricing source files, compiled IPIR AST models, semantic diff reports, and execution traces in gs://rateguard-ai-artifacts.

Pub/Sub Worker Queue

Async Queue

Delivers asynchronous assurance jobs via authenticated push subscription to Cloud Run worker endpoints for background execution.