I design and build the AI systems enterprises actually put in front of regulators, auditors and customers.

Enterprise AI that survives an audit

I build enterprise AI systems in the insurance industry: multi-agent orchestration, RAG pipelines and human-in-the-loop workflows on Azure. Observability, security and governance from day one, with a strong preference for keyless architectures that remove secret sprawl at the infrastructure level.

• Multi-agent systems

  Microsoft Agent Framework and LangGraph, orchestrating specialist agents for document processing, decision support and process automation.

• Agentic UIs

  AG-UI and CopilotKit. Conversational interfaces that expose agent state and reasoning to the person using them.

• RAG pipelines

  Azure AI Search and OpenAI, with retrieval evaluation, grounding verification and content safety built in from the start.

• Reference architectures

  Cloud-native .NET Aspire with Angular or Next.js, following twelve-factor and microservice principles.

Reference architectures, not slideware

Each of these exists because a real project needed it first. All are public and all run.

• E-Commerce Agents

  A multi-agent e-commerce platform in Python and .NET: orchestrator plus five specialists over A2A, with MCP tools, pgvector retrieval and OpenTelemetry throughout. The companion repo for the MAF series.

  Python.NETMicrosoft Agent FrameworkA2AMCP

  26 stars4 forksPythonupdated 13d ago

• AI Knowledge Hub

  A framework-free reference for AI engineering fundamentals: prompting, evaluation, retrieval and agent design, without tying any of it to one SDK.

  MkDocsReference

  1 stars0 forksPythonupdated today

• Microsoft Graph MCP

  An MCP server over Microsoft Graph with on-behalf-of auth, so an agent acts as the signed-in user rather than as an application.

  PythonMCPEntra IDOBO

  4 stars0 forksPythonupdated today

• MCP Generator

  Scaffolds a typed MCP server from an OpenAPI document, so an existing REST API becomes agent tooling without hand-writing the bridge.

  TypeScriptMCPOpenAPI

  1 stars0 forksTypeScriptupdated 2d ago

All 12 projects

The work, written down

Long-form deep dives with the source repository attached. Most run as series, because a production architecture rarely fits in one post.

• MAF v1: Python and .NET29 parts
• Building Multi-Agent AI Systems12 parts
• Clean Architecture: Contact Management Application12 parts
• Defense in Depth for AI Agents3 parts
• Zero-Secrets Azure2 parts
• Build Your Own MCP Server1 part

Read the writing index

• Keyless by default

  Managed identity over API keys, workload identity federation over stored credentials. If a secret exists, it is a place the system can fail.

• Governance is a build task

  Content safety, token quotas, audit trails and identity on both sides of every call, designed in at the start rather than retrofitted after the first incident.

• Observability before scale

  OpenTelemetry traces through every agent, tool and model call. If you cannot see what an agent did, you cannot operate it.

• Write it down with the source

  Every architecture I publish comes with a repository you can run. An architecture nobody can reproduce is an opinion.

Enterprise AI

Microsoft Agent Framework, LangGraph, Azure OpenAI, Microsoft Foundry, Azure AI Search, RAG, Document Intelligence, Anthropic Claude, AI governance and content safety

Azure platform

AKS, Service Bus, Key Vault, App Service, Functions, Private endpoints, Docker, Kubernetes, Helm, Azure Container Registry

Identity and security

Managed identity, Workload identity federation, Zero-secret pipelines, Entra ID service principals, RBAC, Zero trust, OWASP, Azure Policy, Defender for Cloud, Multi-tenant data isolation

Backend and data

C# / .NET, .NET Aspire, Python, MediatR, FluentValidation, Azure SQL, Cosmos DB, PostgreSQL, MongoDB

Frontend

Angular, Next.js, TypeScript, Tailwind CSS

Delivery and observability

OpenTelemetry, GitHub Actions, Azure Monitor, Application Insights, Distributed tracing, Serilog, Azure DevOps