Senior AI Engineer / Architect @ Microsoft
Designing enterprise grade, event driven AI systems that integrate autonomous agents, tools, and self healing workflows into production environments.

stochasticcoder.com • linkedin.com/in/jonathanscholtes/

These reference architectures demonstrate how to move past basic prompt and response into structured, tool driven execution and deterministic agent orchestration.

• Azure SRE Agent GitHub Demo
  Building self healing pipelines with Azure SRE Agent and GitHub Copilot. Showcases automated incident detection and governance, taking a failure spike down to an automated code fix and a merged GitHub PR.

  • Key Patterns: Agentic reliability, autonomous remediation loops, GitHub Copilot integration.

• Azure AI Foundry Agentic Workshop
  A comprehensive, hands on workshop repository covering vector search, multi agent orchestration via LangGraph, and evaluation setups.

  • Key Patterns: Developer enablement, multi agent evaluation frameworks, LangGraph orchestration.

• BrandSense (Multi Agent Brand Intelligence)
  Brand analysis pipeline combining retrieval, scoring, and validation. Applies agent based workflows to business evaluation scenarios to ensure high data integrity.

  • Key Patterns: Guardrailed multi agent collaboration, deterministic validation loops.

• Contract Risk Analysis (MCP + Foundry)
  Contract analysis using MCP for tool based evaluation and data access. Enables structured, repeatable, and auditable analysis workflows.

  • Key Patterns: Tool based evaluation isolation, auditable decision trees.

• ITSM Multi Agent System (Microsoft Foundry)
  IT service management implemented with agents and structured orchestration. Covers ticket classification, routing, and lifecycle handling.

  • Key Patterns: State machine orchestration, autonomous ticket remediation.

• Agents Audit System
  Observability and evaluation framework for agent and tool interactions. Tracks execution flow, decisions, and tool usage across workflows.

  • Key Patterns: Agentic tracing, runtime evaluation, LLM as a judge observability.

The core infrastructure, gateway routing, and scalable processing patterns required to back robust AI platforms.

• MCP YARP Gateway: Gateway layer for routing and controlling tool access in agentic systems. Enables structured interaction between agents and external tools.
• Travel AI Agent (React, FastAPI, Cosmos DB): End to end implementation of an operational agent handling user queries, state management, and transactional bookings against vector databases.
• Large Document Summarization (Durable Functions): Distributed fan out and fan in processing for large scale workflows handling extensive data payloads.
• Azure AI Foundry Deployment: Enterprise ready infrastructure including networking, security, and IaC templates.
• Full Stack RAG System: Retrieval and generation pattern built with React, FastAPI, and Cosmos DB.
• Semantic Kernel RAG System: RAG and orchestration pattern using Semantic Kernel and Azure services. Serves as the foundation for agent based system design.

I write regularly about the operational realities of AI engineering, focusing on reliability, session affinity, and self healing pipelines over speculation.

👉 Read the latest technical breakdowns at stochasticcoder.com

📌 Core Engineering Philosophy

• System Design > Isolated Prompts: Prompts are brittle; architectures must be resilient.
• Governed Autonomy: Agents must operate within explicit operational guardrails.
• Consistent Operational Outcomes: Observability and reliable tracing are non negotiable for production agent deployment.

The architecture patterns above leverage these core Microsoft frameworks and concepts to build scalable, production ready systems:

• Microsoft Agent Framework Journey: The architectural pathway for transitioning from basic AI capabilities to governed multi agent orchestration.
• Microsoft Foundry Planning: Core concepts for designing structured, tool driven execution and deterministic agent planning loops.
• Microsoft Foundry Observability: Foundational practices for tracing execution flow, evaluating decisions, and driving consistent operational outcomes across AI applications.