Mitchel Adiele

AI automation · systems operations

I turn operational problems into systems that run.

I design, direct and operate production workflows across customer access, payments, real-time media, content publishing and AI-assisted operations. My work is measured by what keeps running when the happy path ends.

View four systemsmitcheladiele@gmail.com ↗
300+paying customers supported
1,400+authenticated sessions
2,500recorded viewing sessions
24/7production environment

Selected systems

Four systems running in production.

Each began with an operational constraint and became a maintained workflow.

01 / CUSTOMER OPERATIONS

Subscription Operations Bot

A customer lifecycle system that connects payment confirmation, subscription state, controlled access, reminders and human support.

Payment event→State update→Access policy→Customer action
  • Processes payment webhooks and reconciles successful transactions with subscriber records.
  • Issues controlled invitations, enforces channel access and removes expired memberships.
  • Schedules renewal reminders, sends payment alerts and escalates exceptions for human review.
  • Coordinates six access pools and has supported more than 800 consumed invitations.
02 / REAL-TIME ORCHESTRATION

Media Operations Agent

A decision and control layer for parallel OBS environments. It assigns work, prepares browser media, verifies playback and recovers failures while preserving operator control.

Decision layerschedule · source · ownership
OBS 01OBS 02OBS 03OBS 04
  • Controls isolated OBS instances through dedicated WebSocket endpoints.
  • Uses per-instance Chrome DevTools Protocol ports to inspect browser targets, trigger playback and verify loaded state.
  • Persists channel ownership and active state across service restarts to prevent conflicting control.
  • Monitors playback health, applies source-specific recovery and escalates when automation cannot establish a safe state.
03 / CONTENT AUTOMATION

Real-Time Content Engine

An event-driven agent that turns changing sports data into timely, branded publishing decisions without treating any single provider response as unquestionable truth.

Provider data→Evidence checks→State machine→Publish / edit
  • Tracks fixture lifecycle, lineup changes, live events, terminal states and post-match media.
  • Combines timing gates, source confidence, refresh jobs and forward-only state transitions.
  • Generates visual assets, publishes to Telegram, edits prior output and prevents duplicate delivery.
  • Runs new decision logic in shadow mode and compares proposed actions with production outcomes before authority is expanded.
04 / AGENTIC WORKSPACE

RoseX

A persistent operations workspace that lets me direct an AI agent through Telegram with access to live context, durable memory, files, browsers and operational tools.

  • Separates stable instructions, current state and durable memory so long-running work remains recoverable.
  • Supports mid-task steering, multi-file input, browser automation, MCP tools, artifacts and production actions.
  • Recovers context after server restarts and unexpected process interruption.
  • Keeps human authority explicit for production risk and irreversible actions.
Open the RoseX case study ↗

Implementation language

How I structure automation.

01

Start with state

Define what is true, who owns it and how it survives interruption.

02

Design the failure path

Retries, idempotency, health checks, fallback behavior and escalation are part of the workflow.

03

Make decisions observable

Logs, status APIs, dashboards and operator alerts expose what the system believes and why.

04

Keep a human boundary

Automation handles routine execution. Ambiguous or high-risk states retain a clear override.

APIs & webhooksTelegram automationn8nClaude & CodexPlaywrightOBS WebSocketChrome DevTools ProtocolSQLiteCloudflareGitHubOperational dashboards