Client & Edge

Web/SPA, Mobile Apps, CDN, WAF, DDoS Protection, Geo-routing

API Gateway

Auth, Rate Limiting, Load Balancing, Protocol Translation, Request Routing

Service Mesh

Search, Graph, AI/ML, Auth, Notifications, Sync, Media Processing

Data & Storage

PostgreSQL, MongoDB, Redis, Elasticsearch, Vector DB, S3/CDN

Core Components

Aevum's architecture is built around loosely coupled, horizontally scalable microservices. Each component is independently deployable, monitored, and versioned.

🔍

Semantic Search Engine

Hybrid retrieval combining lexical BM25 scoring with dense vector embeddings for contextual understanding.

  • Multilingual query parsing & normalization
  • Real-time re-ranking via cross-encoder models
  • Faceted filtering & knowledge graph traversal
🧠

AI Verification Pipeline

Multi-stage RAG pipeline that cross-references claims against primary sources and academic databases.

  • Source extraction & citation mapping
  • Confidence scoring & hallucination detection
  • Expert review routing & version control
🌐

Knowledge Graph Engine

Property graph database maintaining entities, relationships, and temporal metadata across disciplines.

  • Ontology alignment & schema evolution
  • Transitive closure & path reasoning
  • Visual query builder & export APIs
🔄

Event Sync & CQRS

Command Query Responsibility Segregation ensuring low-latency reads while maintaining ACID compliance on writes.

  • Kafka-backed event streaming
  • Materialized views & projection updates
  • Conflict resolution & optimistic locking

Request Data Flow

When a user submits a query, the system routes through an optimized path balancing latency, accuracy, and cache efficiency.

1. Client → Edge CDN (Cache Check) IF hit: return cached response (TTFB < 50ms) ELSE: forward to API Gateway 2. API Gateway → Auth & Rate Limit Validate JWT / API Key Apply tier-based throttling 3. Gateway → Search Service Parse query → normalize → translate (if needed) Execute hybrid search (BM25 + Vector) 4. Search → Verification Layer Fetch top-K candidates Run cross-encoder re-ranking + fact-checker 5. Response Assembly Merge metadata, citations, graph edges Cache result → return JSON/GraphQL

Technology Stack

Curated for reliability, performance, and developer velocity. All dependencies are pinned and automatically audited.

Category Technology Role Status
Runtime Node.js 20 LTS / Rust (perf-critical) Service Execution Production
Framework NestJS, Actix-web, tRPC API & Service Orchestration Production
Databases PostgreSQL 15, MongoDB 7, Redis 7 Relational, Document, Cache Production
Vector/Graph Weaviate, Neo4j, pgvector Semantic Search, Knowledge Graph Active Dev
ML/AI PyTorch, ONNX, vLLM, LangChain RAG, Embeddings, Verification Production
Infrastructure Kubernetes, Terraform, AWS/GCP Orchestration, IaC, Cloud Production

Performance & SLA

Global infrastructure optimized for sub-100ms read latency and 99.99% availability across 140+ regions.

99.99%
Uptime SLA
<65ms
Avg Read Latency (p95)
42K+
Requests/Second
94.8%
Cache Hit Ratio
3
Active Regions
<2min
Failover Recovery