AI Agent Architect

Adam Matthew Steinberger Portfolio

Staff Software Architect & AI Automation Engineer

Built vibey, an open-source, PostgreSQL-backed conductor that runs five interchangeable coding-agent engines (Claude, Codex, Cursor, Gemini/Antigravity, Qwen) through one queue with enforced per-project cost caps — 1,440+ commits on GitHub and dogfooded to build the very site presenting it.

AI AgentsPythonAzureRAGKubernetes

Location

Greenville, SC

Companies

The Vizius Group

Skills

AI AgentsPythonAzureRAGKubernetes

Notable

  • the-vibey-project/vibey on GitHub: 1,440 commits, 17 tagged releases, distributed as one PyPI package (uv tool install vibey)
  • Author of "Novice to Navigator: Your Guide to AI Chatbots for Business" — first edition free to read, second edition in development
  • GitHub profile independently corroborates the portfolio's bio, title, and location, and lists an ORCID researcher ID (0009-0007-6853-8981)
  • 13+ years of Azure experience, including advisory work on identity governance for a SOX-regulated enterprise with ~5,700 workforce identities

What makes it work

A breakdown of the choices that make this portfolio stand out.

1

CLI walkthrough captioned with reasoning, not just syntax

The "Run the whole stack in ten minutes" section is six real commands (uv tool install vibey, vibey new, vibey doctor --conformance --record, vibey worker --provider claudeloop --engines claudeloop,agyloop -j 2...), each with a one-line caption explaining why the flag exists — "a real budget brake, enforced from the ledger" next to --max-cycle-dollars, "every database-backed command reads it; vibey never guesses a database" next to the Postgres URL. That is design reasoning exposed inline, not a generic quickstart.

2

Case studies labeled by role and led with hard numbers

Each of the five "Who's Behind It" cards carries an explicit role tag (SOLE ARCHITECT, CO-LEAD, SOLE AUTHOR, LEAD) and opens with a quantified claim instead of a summary sentence: "20 microservices, 585 test modules, human approval on every phase," "40 resource kinds reconciled from Git," "653 tests, 93% coverage, chaos-proved convergence." No adjectives doing the work that numbers should.

3

Three distinct audiences, three distinct proof artifacts

The "Who This Is For" section splits developers, governments/military, and universities into separate pitches, each pointing at a different kind of evidence: a contribution guide for developers, an audit-trail/local-model argument for government reviewers, and a formal paper plus a full documentation book for academics. That is deliberate segmentation, not one generic "hire me" page reused three ways.

4

The open-source claim is independently checkable, not just asserted

the-vibey-project/vibey on GitHub shows 1,440 commits, 17 tagged releases, and an active develop branch with a recent chore commit — not a single abandoned demo repo. The personal GitHub profile (adammatthewsteinberger) independently corroborates the same bio, title, and location shown on the portfolio site, and links back to it.

What AI Agent Architects can take from this

Specific, actionable tips to apply to your own portfolio — no generic advice.

  • Caption every CLI or code snippet with the design reasoning behind it, not just what the command does — "a real budget brake, enforced from the ledger" tells a reviewer more than the flag name alone ever could.

  • Label each case study with your actual role on it (Sole Architect, Co-Lead, Lead) and open with a hard number — test count, service count, coverage percentage — before any summary sentence.

  • If your work genuinely serves different audiences, give each one its own section and its own proof artifact instead of one generic contact form.

  • Ship a real, versioned open-source package with tagged releases rather than a single unstructured repo — a visible release history (17 tags, 1,440 commits) is a credibility signal a recruiter can check without asking you.

  • State your documentation habit as a concrete analogy tied to your own domain ("I document everything for the same reason a RAG pipeline cites its sources") instead of the generic "I value documentation."

Case studies

An open-source agent conductor, the enterprise gateway work behind it, and a plain-English primer for the buyers. The same concerns run through all three: budgets, audit trails, and a human at every decision that matters.

Case StudyadvancedVizius engagement, 2026

AI Governance Gateway

PythonFastAPIAzure OpenAI / FoundryRedisAzure Service BusEntra ID workload identityPrivate AKSKEDAMCPNext.js 15

A central gateway that puts Azure AI services plus Anthropic, OpenAI/Codex, Cursor, xAI Grok and Gemini behind one OpenAI-compatible client. Every call goes through the same pipeline: resolve connection, evaluate policy, rate-limit, invoke, compute cost, write a usage record. Workloads authenticate with Entra ID tokens, so there are no API keys anywhere in the path.

It is agent governance in a real enterprise: cost caps as a denial-of-wallet control, a tamper-evident audit trail, and an OpenAI-compatible surface so migrating takes a two-line change.

  • Five model vendors behind one policy-enforced, OpenAI-compatible API

  • Multi-unit rate limiting in Redis and per-call USD cost caps

  • Zero stored credentials: Entra ID tokens and federated workload identity

Case StudybeginnerFirst edition free online; second edition in development

Novice to Navigator: Your Guide to AI Chatbots for Business

Long-form articlesFeynman TechniqueReadiness quiz

A 33-article series in 7 sections that takes a non-technical business reader from 'what is AI' through chatbots, RAG and advanced concepts to evaluating and hiring someone to build a custom solution. It is written using the Feynman Technique, with each section building on the last.

An architect who can explain RAG and prompt engineering to the people signing the budget is rare, and the series is a working example.

  • 33 articles across 7 progressive sections

  • Plain-language treatment of RAG, prompt engineering and chatbot ROI

  • Free to read, with a readiness quiz for buyers

Projects

Open SourceadvancedOngoing; 1,513 commits as of Sep 2026

vibey

Python 3.12PostgreSQLClaude CodeOpenAI CodexCursor AgentGemini/AntigravityQwenPyPI

A queue-based, six-phase conductor for autonomous software delivery. It interviews you until the spec is clear, builds unattended across a pool of coding-agent engines, and brings you back only for decisions that are yours to make. Every choice, finding and handoff goes into an append-only PostgreSQL ledger rather than one vendor's chat session.

It handles the unglamorous parts of agent delivery (crash recovery, budget brakes that are actually enforced, vendor failover without losing open questions) with a real database behind them.

  • Engine pool across Claude Code, Codex, Cursor, Antigravity and local Qwen/GPT-OSS runners

  • Per-project dollar caps enforced from the ledger (--max-cycle-dollars)

  • PostgreSQL queue with FOR UPDATE SKIP LOCKED; survives crashes and credit exhaustion

Ready to build your portfolio?

Follow the AI Agent Architect roadmap — skills, projects, and timeline to get hired.

AI Agent Architect Roadmap