Case studies

What he built, decided, owned and proved.

Three hiring-relevant cases first. The complete project inventory remains available below without dominating the story.

01

Knowledge operations

Optus-scale proof
Scale + adoption

Knowledge and support operations at national frontline scale

Problem

Duplicated processes, support load and search friction across a large frontline audience.

System

Process redesign, knowledge maintenance, content removal/rewrite, training, communications and reporting.

Decision

Treat documentation quality, ownership, staff adoption and measurement as one operating loop.

Ownership

Operational analysis, process intervention, rollout support and performance visibility.

Proof

352 stores · 5,000+ employees · 500,000 monthly knowledge-base visits · 74% recorded output increase over 12 months.

See chronology and evidence
02

Business systems

Current ownership
Commercial + operational

Multi-brand AI and workflow operating layer

Problem

Lead flow, quote preparation, customer communication, admin, reporting and field handoffs spread across brands and tools.

System

Websites, lead capture, routing, qualification, follow-up, exception dashboards, internal tools and bounded agent workflows.

Decision

Automate repeatable low-risk work; keep named human approval for money, compliance, quality and customer commitments.

Ownership

Discovery through deployment, observability, governance, documentation and staff handover.

Proof

Current role and system map are inspectable. Commercial outcome metrics are private and not published here.

Open operating map
03

Agent operations

Governed runtime
Orchestration + control

Visible, permissioned multi-agent and model-routing operations

Problem

Agents, models, accounts and machines create invisible work, auth uncertainty, duplicated context and permission risk.

System

Hermes, Codex/APIs, Claude, local LLM routing, shared state, schedules, watchdogs, usage tracking and operator receipts.

Decision

Route by capability, privacy, latency and cost; fail closed for public, financial and destructive actions.

Ownership

Architecture, implementation, safety model, acceptance tests, runtime monitoring and recovery paths.

Proof

Public repos, architecture artifacts and screenshots. Private runtime evidence remains labelled; throughput is not presented as a business outcome.

PROJECT ARCHIVEOpen 12 systems and artifacts
Project archive

12 systems and artifacts

michaelcostea.com / MICHAEL OS 89 project artifact
01
Live portfolio

michaelcostea.com / MICHAEL OS 89

The live portfolio operating system itself: a Win95/Nous-zine showcase for AI education, project proof, demos, deck previews, intake, and repo cards, deployed through GitHub Pages.

Scope: Information architecture, responsive desktop shell, résumé and project surfaces, analytics, print assets, regression coverage, browser QA, and GitHub Pages deployment.

Stack: Semantic HTML · CSS · vanilla JavaScript · GitHub Pages · Node.js/jsdom regression tests · Playwright browser QA

portfolio · GitHub Pages · MICHAEL OS 89 · proof hub
RebateSignal project artifact
02
Guarded staging

RebateSignal

Live guarded Victoria-first quote service for source-backed hot-water VEECs, solar and battery STCs, approved-product evidence, and fail-closed unsupported paths.

Scope: Source ingestion and normalisation, parity-checked quote engines, promoted SQLite register releases, authenticated quote workspaces, readiness gates, evidence PDFs, and guarded Firebase staging.

Stack: Python 3.11 · FastAPI · Pydantic · pytest · TypeScript · Node.js 22 · Express · Firebase Functions/Hosting/Firestore/Storage · SQLite · PDFKit

VEU/VEEC · STC · Solar Victoria · rules engine
InvoicePipe project artifact
03
Private system

InvoicePipe

Invoice automation system with OCR/media smoke tests and a commercial private mirror. The showcase screenshot captures the product surface while repo access stays private.

Scope: Supplier-email and upload intake, Australian invoice schema validation, OCR extraction, preview, authentication, database persistence, health checks, and Excel/SharePoint/JSON/API export paths.

Stack: Next.js 15 · React 19 · TypeScript · Prisma · PostgreSQL · NextAuth · Python/FastAPI · Azure Content Understanding · Power Automate

invoice automation · OCR · document ops
Brief2Ship project artifact
04
Public repo

Brief2Ship

Public operating standard for turning rough AI requests into better first-pass builds, reports, and handoffs: one strong brief, the right existing base, a maintainability gate, anti-slop finish, and proof receipts.

Scope: Hermes/OpenClaw skill packaging, app/dashboard/landing/report lanes, kickoff and receipt templates, maintainability and visual-finish gates, documentation validation, and one-shot brief-checker dogfood.

Stack: Markdown · YAML-frontmatter Hermes/OpenClaw skill · Python validation · shell installer · HTML/CSS/JavaScript demo · Git/GitHub

AI build standard · report lane · proof receipts · repo-first
Automated Social & Brand Content Engine project artifact
05
Live internal workflow

Automated Social & Brand Content Engine

End-to-end research-to-publish operating system for personal-brand and channel content: monitors AI and repository signals, creates draft calendars, packages LinkedIn activity and short-form video, and feeds performance results back into planning.

Scope: Industry and repository monitoring, research and draft queues, content calendars, media assembly, Shorts metadata QA, review and approval packages, publishing handoffs, Telegram receipts, and analytics feedback loops.

Stack: n8n · Python · JSON/CSV state and ledgers · HTML/CSS/JavaScript dashboards · local LLM and agent harnesses · LinkedIn, YouTube, TikTok, Snapchat, Instagram, and Patreon workflow lanes

content operations · LinkedIn · YouTube Shorts · n8n · QA gates · analytics
UseAIForMe.com project artifact
06
Live site · private repo

UseAIForMe.com

Early-access AI operator marketplace and consultancy surface: buyers post a real outcome, operators show proof and pricing, and paid work starts only after scope, fit, and approval are confirmed.

Scope: Responsive marketplace UX, buyer and operator journeys, outcome intake, proof and package surfaces, Express service layer, Firebase deployment, and approval-first Stripe checkout.

Stack: React 19 · TypeScript 6 · Vite 8 · Express 5 · Firebase Hosting/Functions/Admin · Stripe · Vitest · jsdom · Testing Library

marketplace · AI operators · approval-first checkout
Telegram Office / Agent Office project artifact
07
Public repo · runtime hidden

Telegram Office / Agent Office

Pixel office for visible multi-agent work: Telegram-triggered events, bot rosters, rooms, handoffs, live activity zones, and browser-first operator visibility.

Scope: Telegram webhook and Hermes gateway ingestion, roster and room state, lifecycle and activity telemetry, context health, attention queues, multi-machine receipts, and an auth-protected operator surface.

Stack: Next.js · React · TypeScript · Phaser · Telegram webhooks · Hermes gateway logs and telemetry · automated TypeScript/browser test suite

agent visibility · Telegram · pixel office
Agentic Framework Session project artifact
08
Inspectable artifact

Agentic Framework Session

Education and enablement deck explaining Hermes, safe agent setup, shared memory, Discord/team rooms, review gates, and how businesses can use agents without silent risk.

Scope: Interactive browser presentation and downloadable PDF covering agent concepts, Hermes setup, model access, shared memory, Discord operating rooms, permissions, review gates, and practical business examples.

Stack: Semantic HTML · CSS · inline SVG diagrams · vanilla JavaScript · Node.js build scripts · Playwright/Chromium PDF rendering

AI education · agent safety · shared memory
AI Profile Sites project artifact
09
Public repo

AI Profile Sites

n8n/Hermes profile-site generator that turns structured person/company inputs, evidence, screenshots, and source receipts into polished profile pages for clients, founders, and operators.

Scope: Structured person/company intake, Hermes research jobs, hold-and-resume recovery, source receipts, generated responsive pages, desktop/mobile previews, and automated GitHub Pages publishing.

Stack: n8n · Python 3 · Hermes Agent · HTML/CSS · JSON state and source receipts · Git · GitHub Pages

profile generator · n8n · Hermes · source receipts
AgentMesh / Multi-Agent Control Kit project artifact
10
Local-first kit

AgentMesh / Multi-Agent Control Kit

Local-first coordination package for teams of AI agents: shared mailboxes, relay demos, handoffs, visible receipts, bootstrap docs, security notes, and public-alpha packaging.

Scope: Machine and agent discovery, transport registries, six local transports, append-only project and activity state, relay/mailbox dispatch, bootstrap tooling, security notes, and approval-gated system changes.

Stack: Python CLI · JSONL state stores · shell/bootstrap scripts · Python packaging and wheel builds · pytest · Hermes/OpenClaw/Claude Code/Codex local transports

multi-agent control · mailbox · handoffs · receipts
Codex Account Usage + Auth Rotator project artifact
11
Private operator tool

Codex Account Usage + Auth Rotator

Private operator tool for tracking Codex account usage, reset windows, active account status, warmup, switching, and auth rotation across a pool without guessing which login is safe to use.

Scope: Profile import, browser PKCE and device OAuth, usage polling, five-hour and seven-day windows, warmup, account scoring and switching, safe auth-error diagnosis, and a loopback dashboard.

Stack: Rust 2024 · Tokio · Ratatui · Crossterm · Reqwest · Serde · OpenAI OAuth/usage APIs · Python loopback service · vanilla JavaScript dashboard

Codex usage · auth rotation · account pool · operator tooling
12
Public repos

Repo-First Starter + Cursor Covenant

Two small public operator-safety repos: one pushes AI builders to discover the best existing base before greenfield work; the other adds a visible cursor-control countdown and hard UI-control rule.

Scope: Repo/tool discovery and scoring, local-workspace search, maintainability entropy checks, portable agent instructions, install scripts, plus visible cursor leases with countdown, cancel, and narrowly scoped UI control.

Stack: Python standard-library CLI · GitHub REST API · pytest · Markdown/YAML agent skills · Bash installers · PowerShell/WPF · portable AGENTS/CLAUDE/Cursor rules

agent safety · repo discovery · UI control
Need the short version?

Download the two-page résumé.

INTERACTIVE PORTFOLIOMichael OS 89