Reference Architectures & Case Studies

Architecture built to production standards,
designed to be governed

Seven architecture builds covering enterprise RAG, multi-agent systems, data platforms, and AI governance. Three of them form a deliberate vertical stack: an organisational governance layer, a live multi-agent application with human oversight, and an independent evidence layer that evaluates the system against NIST AI RMF and OSFI E-23 controls — built to the standard a regulated enterprise would require.

Organisational Layer
AI Governance Maturity Assessment
25-question tool scoring AI readiness across five control domains. Live at assessment.cloudkraft.com.
Application Layer
Regulatory Intelligence Agent
Live multi-agent system with HITL governance, append-only audit log, and LangSmith tracing. See featured build below.
Evidence Layer
Enterprise AI Assurance Framework
RAGAS evaluation suite with 15-item golden dataset. Generates audit-ready Model Risk Assessments mapped to NIST AI RMF and OSFI E-23.
Latest · Live Deployment
Agent Systems · Governance
Regulatory Intelligence Agent — Governed Multi-Agent System
A LangGraph multi-agent system that answers compliance questions by retrieving grounded regulatory context, drafting cited responses, and proposing audited write actions — all under mandatory human oversight. Live at reg-intel.demo.cloudkraft.com. Ships with an independent Enterprise AI Assurance Framework: RAGAS evaluation suite, 15-item golden dataset, and a Model Risk Assessment report mapped to NIST AI RMF and OSFI E-23. Production path to AWS is fully specified in five Architecture Decision Records.
LangGraph Multi-Agent FastAPI pgvector RAG Governance HITL RAGAS Evaluation AWS Bedrock Path Railway
Read Case Study →
Live Demo ↗
01
RAG · Knowledge Architecture
Building the Institutional Brain — Enterprise Knowledge Navigator
A five-layer RAG reference architecture combining MCP server integration, Weaviate vector store, and provider-agnostic LLM gateway. Non-agentic by design — every answer is cited and auditable, with the human firmly in the decision seat.
LangChain Weaviate MCP RAG Governance Next.js 14
Read Case Study →
02
Architecture Deep Dive
Why I Built a Framework, Not Bespoke AI Systems for Each Use Case
A behind-the-scenes look at the reusable Enterprise RAG Framework I built to codify how enterprise AI deployments should be approached — and why the architecture decisions that seem over-engineered are the ones that matter most in production
RAG GenAI Architecture Enterprise
Read Case Study →
03
Agent Systems · Portfolio Demo
EA Advisor Agent — Four AI Agents, One Architecture Report
A deployed multi-agent system built with CrewAI where four GPT-4o-mini specialists — business analyst, solution architect, risk assessor, cost optimiser — run sequentially over the same requirements and assemble a unified advisory report with a live C4 architecture diagram. Engineered around an async job pattern to eliminate proxy timeouts on long-running LLM chains.
CrewAI Multi-Agent FastAPI GPT-4o-mini Next.js 14 OpenRouter
Read Case Study →
Live Demo ↗
04
Platform Architecture · ADR-Governed
Enterprise AI Gateway
Hybrid inference platform with local Ollama (primary) + OpenRouter fallback, semantic Redis caching, PostgreSQL audit persistence, and three documented ADRs. Demonstrates how to balance privacy, cost, and resilience through explicit routing policy and governance-first design.
FastAPI PostgreSQL Hybrid Routing Ollama Redis ADR-Documented
Read Case Study →
Live Demo ↗
05
Integration Pattern · Orchestration-First
Enterprise AI Integration Pattern — Your AI Works. Nothing Connects to Anything.
An event-driven workflow orchestration layer that connects document intelligence to the systems where decisions actually get made — the architecture pattern for the most common failure mode in enterprise AI: accurate models that never reach a real workflow. Six-layer platform core configured per use case, not rebuilt for it.
n8n FastAPI Qdrant Event-Driven
Read Case Study →
06
Data Strategy · Reference Architecture
Scalable Architecture for Real-Time Analytics
Reference architecture for a large omnichannel enterprise — unifying fragmented operational systems, eliminating manual data movement, and enabling real-time finance and HR analytics on AWS. Documents the data platform design decisions that underpin AI-readiness.
AWS Data Lake Data Governance Real-Time Analytics
Read Case Study →
07
Live Commercial Tool · CloudKraft AI Control Framework™
AI Governance Maturity Assessment
A 25-question interactive assessment scoring your organisation across five AI Control Framework layers — Data Readiness, Architecture Patterns, Governance & Risk, Operating Model, and Integration & Scale. Produces a scored maturity profile, gap analysis, and prioritised executive roadmap in under eight minutes. No account required. Used to qualify organisations for Tier 1–3 CloudKraft advisory engagements.
AI Governance Maturity Assessment CloudKraft AI Control Framework™ 25 Questions No Account Required Executive Roadmap Output
Launch Assessment ↗
CloudKraft Consulting Inc.
Need this kind of architecture thinking for your organisation?
Engagements start with a structured diagnostic — identifying where AI can create genuine leverage in your current systems and what architecture decisions need to be made first.
View Consulting Services