Portrait of Bob Tryon

J. Robert Tryon, Jr.

Agentic AI Architect.NET Enterprise ArchitectDistributed SystemsAzure Cloud

Media Six LLCLas Vegas, NV844-400-5250bob@mediasix.co

Media Six AI

Agentic AI Engineering — building Agentic AI solutions with Microsoft Agent Framework

Microsoft Agent FrameworkLangChainCrewAIRAGMCP

  • CalTech AI/ML Certified
  • Microsoft Applied Agentic AI (in progress)

Featured AI Projects

Agent frameworks, tooling, and applied AI systems. The AI projects are my most recent work, built on many years of C# and .NET engineering. Most of them are written in C#, and many run on Microsoft Agent Framework.

Agent Orchestration Tutorials

Open Source

A 24-lesson C# series on Microsoft Agent Framework, the successor to Semantic Kernel and AutoGen. It builds from a single governed agent with tools, MCP, middleware, and approvals to graph workflows using the sequential, concurrent, handoff, group chat, and Magentic-One patterns. The capstone is a checkpointed Researcher → Analyst → Writer pipeline with a human-approval step, hosted on Foundry.

  • Microsoft Agent Framework
  • Agents
  • C#
  • Tutorial

Agent Skills Tutorials

Open Source

A two-lesson C# series on Agent Skills in Microsoft Agent Framework: portable packages of instructions, reference documents, and scripts in the open agentskills.io format. Agents see only each skill's name and description and load the rest on demand, so they carry many skills without bloating the prompt. The series goes from one file-based skill to a single agent composing file, class, and inline skills.

  • Microsoft Agent Framework
  • Agent Skills
  • C#
  • Tutorial

MiniCode - Coding Agent Tutorial

Open Source

A 28-lesson course that builds MiniCode, a lightweight Claude Code–style CLI coding agent in C# on Microsoft Agent Framework. A single agent drives the whole task through a tool loop: it explores the repository, plans before editing, makes sandboxed file edits, runs dotnet build and test, then diagnoses and repairs its own failures. Git integration and human approval for destructive operations round out the capstone.

  • Microsoft Agent Framework
  • Coding Agent
  • C#
  • Tutorial

Semantic Kernel Tutorials

Open Source

A day-by-day C# series on Microsoft Semantic Kernel with OpenAI, where every lesson is a standalone, runnable project. Part 1's 30 lessons run from a single LLM call through tool calling, RAG, a hand-rolled ReAct loop, semantic routing, telemetry, and LLM-as-a-judge evaluation, ending in a GoalRefiner → Planner → Developer → Reviewer capstone deployed as a Web API. Part 2, in progress, closes the remaining orchestration-pattern gaps: self-reflection, voting ensembles, group-chat debate, market negotiation, plan-and-execute, guardrails, and episodic memory.

  • Semantic Kernel
  • Orchestration
  • C#
  • .NET
  • Tutorial

Skills

Claude Code skills I have built to package repeatable engineering work: scaffolding Blazor apps to a fixed architecture, writing technical courses, and tracing bugs to their root cause.

Skill Tags

Claude Skills - Blazor Studio

A Claude Code skill that makes Claude a principal Blazor and C# architect, so the same requirements produce the same architecture every time. It scaffolds a .slnx solution split into four n-tier projects: Models, DataServices, FunctionalServices, and a Blazor Web App. From a new table or an existing database schema, it generates SQL scripts, EF Core migrations, partial repositories with CRUD and per-index queries, and paged list, edit, and read-only pages in Tailwind or Bootstrap. It works with EF Core or Dapper over SQL Server, MySQL, PostgreSQL, CosmosDB, or MongoDB, and keeps database credentials out of the prompt.

  • Claude Code
  • Agent Skills
  • Blazor
  • C#

Claude Skills - Tutorial Instructor

A Claude Code skill that makes Claude a curriculum designer and technical instructor for course series, written lessons, and workshops. It first lays out a syllabus in phases, checks that each lesson only builds on what came before, flags grab-bag lessons, and plans toward a capstone that reuses earlier pieces. It then writes one paste-ready lesson at a time: overview, prerequisites, setup, core concepts, a complete runnable walkthrough, explanation, and expected result. For fast-moving frameworks it checks current package names and APIs instead of trusting memory, and states plainly what it couldn't verify.

  • Claude Code
  • Agent Skills
  • Curriculum
  • Tutorial

Claude Skills - Root Cause Analysis

A Claude Code skill that turns a bug or story ticket into an evidence-backed root-cause analysis. It pulls the ticket, comments, and attachments from Azure DevOps, then works through the repository: searching for exact error strings, following stack traces, reading the code path end to end, and checking git history, tests, and config. It writes a report that separates the proximate cause from the underlying one, cites file and line evidence, and recommends a fix and tests, with optional Mermaid diagrams. Before handing it over, it runs up to three adversarial review passes and lists anything still unresolved.

  • Claude Code
  • Agent Skills
  • Azure DevOps
  • Debugging

Case Studies

Applied AI systems built for real products, from curating a million-diamond catalogue to an agent pipeline that turns assigned tickets into root-cause analyses and planned fixes.

PureIce — AI-Driven Diamond Curation

Case Study

A diamond shopping platform that scores every stone in a million-diamond supplier feed for beauty, then narrows the catalogue to the three best diamonds for each shape and budget. Shoppers start with a short, ranked list instead of thousands of near-identical listings.

  • Applied AI
  • Curation
  • Commerce

DevAssistant — Agentic Ticket-to-Fix Pipeline

Case Study

A desktop multi-agent system that watches a developer's assigned tickets, keeps a root-cause analysis current on each one, and plans the fix in an isolated git worktree. Built on Microsoft Agent Framework workflows with role-scoped Claude agents: each agent's tool set is its capability boundary, and no code is pushed without explicit approval.

  • Microsoft Agent Framework
  • Multi-Agent
  • Claude
  • C#
  • WPF

Systems & Infrastructure

Engineering & Tooling

Developer tooling and lower-level systems work: code generation, protocol clients, terminal emulation, and workflow infrastructure.

  • Blazor Studio

    Visual Studio 2022 extension that generated line-of-business Blazor applications and REST API microservices from a database schema: CRUD pages, list, edit, modal and split views, and an N-tier service layer, with a drag-and-drop form designer and editable templates. Reached 2,300+ installs on the Visual Studio Marketplace. The VS2022 extension is now retired.

  • NodeFlex

    Remote workflow module for n8n, executing workflows against privately hosted infrastructure.

  • TN5250 Emulator

    IBM 5250 terminal emulator with telnet negotiation, EBCDIC translation, and field-level screen rendering.

  • SSH Terminal Emulator

    Terminal emulator over SSH with full VT sequence handling and session management.

    Private repo

Credentials & Learning

Formal training and in-flight coursework.

  1. AI/ML Certification

    Completed

    CalTech November 2024

    • Machine learning foundations, model training and evaluation
    • Deep learning and neural network architectures
    View certificate
  2. Applied Agentic AI Certification

    In progress

    Microsoft In progress — expected December 2026

    • Agent design patterns and multi-agent orchestration

Case Study

PureIce — AI-Driven Diamond Curation

A shopping experience that uses AI to help you find the best diamond for your engagement ring.

  • Applied AI
  • Curation
  • Commerce
  • Recommendation

Problem

A grading certificate describes a diamond but doesn't say how beautiful it is. Two stones with the same carat, color and clarity can look completely different, and shoppers facing thousands of near-identical listings have no reliable way to tell them apart.

Approach

Every diamond in the supplier feed is scored for beauty by a model built for its shape, producing a 0–100 Beauty Index with a seven-part breakdown a shopper can read. Scored stones are then indexed by shape and budget and ranked, so the storefront shows only the top three in each category.

Outcome

One consistent beauty standard applied across a catalogue of over a million diamonds, and a shopping experience that replaces endless filtering with three strong, explained choices. Rankings rebuild on every inventory refresh, so the picks stay current as stones sell.

Case Study

DevAssistant — Agentic Ticket-to-Fix Pipeline

A multi-agent pipeline that carries a work item from assignment to an approval-gated fix, with every agent held to exactly the tools its role needs.

  • Microsoft Agent Framework
  • Multi-Agent
  • Claude
  • C#
  • .NET
  • WPF
  • SQLite

Problem

Every ticket starts with the same groundwork: read the history, find the code, work out why it's broken, and write it down before fixing anything. It's slow, it goes stale the moment the ticket changes, and no coding assistant watches a whole sprint or can prove what an agent was allowed to touch.

Approach

Each ticket moves through a pipeline of Microsoft Agent Framework workflows, from intake and root-cause analysis to planning, coding, review and pull request. Every agent's tool set is its capability boundary: the analyst can't edit source, the test agent can't rewrite tests to make them pass, and only the push agent can reach a remote, and only with human approval. A SQLite ledger advances a stage only when it succeeds, so a crash never leaves half-finished work.

Outcome

In active development. Running against a live Azure DevOps organisation, it adopts assigned tickets automatically and writes a root-cause analysis and implementation plan for each one in its own worktree. Capability boundaries are enforced in code, not prompts, and covered by more than 200 automated tests. Coding, multi-reviewer code review and pull requests land next.