Architecture

Deep dive into the system architecture, design decisions, and technical details.

Overview

This section explains how the NorthBuilt RAG System is architected, why certain technologies were chosen, and how much it costs to operate.

Documents

1. System Architecture

Complete technical architecture with component diagrams and data flow.

  • Component breakdown (Lambda, API Gateway, Bedrock, S3 Vectors)
  • Data flow diagrams
  • Scalability and disaster recovery
  • Performance characteristics

Audience: Engineers, Architects Time: 30 minutes

2. Cost Analysis

Monthly cost breakdown and optimization strategies.

  • Base infrastructure costs (~$137/month)
  • Usage-based costs (Bedrock tokens)
  • Cost optimization strategies
  • Comparison with alternatives
  • Break-even analysis

Audience: Engineering Managers, Finance, Architects Time: 20 minutes

3. Security

Security architecture and compliance.

  • Authentication & authorization
  • Data encryption (at rest and in transit)
  • Network security
  • Incident response procedures
  • Compliance (SOC 2, GDPR)

Audience: Security Engineers, Compliance Time: 25 minutes

4. Architecture Decision Records

Why we made key technical decisions.

  • ADR-001: Serverless on Lambda
  • ADR-010: S3 Vectors for vector storage
  • ADR-011: Lambda Response Streaming
  • Plus 8 more decisions documented

Audience: All engineers, Architects Time: 15 minutes

5. RAG Changelog

History of implemented improvements and optimizations.

Audience: Engineers tracking changes Time: 10 minutes

6. RAG Improvements Backlog

Pending improvements and technical reference.

Audience: Engineers planning work Time: 15 minutes

For New Engineers: System Architecture → Security → ADRs For Cost Analysis: Cost Analysis → ADRs (see alternatives considered) For Security Review: Security → System Architecture → ADRs

Key Takeaways

  • 100% Serverless: No servers to manage, auto-scaling, pay-per-use
  • Cost-Effective: ~$140/month for moderate usage (1K queries)
  • High Performance: <2s end-to-end query latency, <25ms vector search
  • Secure by Design: Multi-layer security, encryption everywhere, IAM least-privilege

Next: Operations Guide to deploy and maintain the system.

Table of contents