Architecture Overview

QuantumVerifi is designed as a set of stateless services backed by managed datastores. This architecture supports horizontal scaling, zero-downtime deployments, and multi-region operation.

Component Diagram

                    ┌──────────────────────┐
                    │    Load Balancer      │
                    │   (Ingress / ALB)     │
                    └──────┬───────┬───────┘
                           │       │
              ┌────────────┘       └────────────┐
              ▼                                  ▼
     ┌─────────────────┐              ┌─────────────────┐
     │   Web Frontend   │              │    API Server    │
     │   (Next.js)      │              │    (Go / Gin)    │
     │   Port 3000      │              │    Port 8080     │
     └─────────────────┘              └────────┬─────────┘
                                               │
                              ┌────────────────┼────────────────┐
                              ▼                ▼                ▼
                    ┌──────────────┐  ┌──────────────┐  ┌──────────────┐
                    │    Worker    │  │   PostgreSQL  │  │    Redis     │
                    │  (Temporal)  │  │              │  │              │
                    └──────┬──────┘  └──────────────┘  └──────────────┘
                           │
              ┌────────────┼────────────┐
              ▼            ▼            ▼
     ┌──────────────┐ ┌──────────┐ ┌──────────────┐
     │  LLM Gateway │ │ Sandbox  │ │   Object     │
     │  (LiteLLM)   │ │  Pods    │ │   Storage    │
     └──────────────┘ └──────────┘ └──────────────┘

Service Details

API Server

The API server handles all HTTP requests — analysis creation, result retrieval, billing, authentication, and SSE progress streaming.

• Stateless — all state is in PostgreSQL and Redis
• Horizontally scalable — run 2+ replicas behind a load balancer
• Authentication — JWT-based with configurable identity provider

Worker

Workers execute analysis workflows using durable orchestration. Each analysis is a multi-step workflow that survives infrastructure restarts.

• Auto-scaled — scale based on queue depth
• Heartbeat monitoring — long-running analyses send periodic heartbeats
• Retry logic — failed steps are retried with exponential backoff

Web Frontend

The Next.js frontend provides the dashboard, analysis viewer, settings, and billing UI.

• Server-side rendering — fast initial page loads
• Real-time updates — SSE streaming for analysis progress

LLM Gateway

Routes LLM requests to configured providers with automatic failover:

• Primary — Azure OpenAI (GPT-4.1 family)
• Secondary — Anthropic (Claude Sonnet / Haiku)
• Tertiary — Ollama (self-hosted, for air-gapped deployments)

Rate limiting, response caching, and cost tracking are handled at this layer.

Sandbox Execution

Tests run in ephemeral containers with per-language runtimes:

Sandboxes are created on demand, run for the duration of the test, and are destroyed immediately after — no persistent state, no shared resources.

Data Flow

1. User submits analysis via API or dashboard
2. API server creates a workflow and enqueues it
3. Worker picks up the workflow, clones the repo
4. Worker calls LLM Gateway for test generation
5. Generated tests are sent to a sandbox for execution
6. Results are stored in PostgreSQL + Object Storage
7. Evidence chain events are hashed and linked
8. SSE events stream progress back to the frontend

Scaling Guidelines

Network Requirements