CHRISTOS GALAIOS

AI Automation Engineer.

I identify operational bottlenecks and eliminate them through purpose-built AI systems —
autonomous pipelines, 110-tool MCP servers, and multi-agent workflows that run without ongoing manual input.

0 Automated Tests

0 MCP Tools

0 Autonomous Codebases

0 Engineers Led

0 Years Experience

[ PROJECTS ]

What I've Built

SocialiseHub

6,300+ tests

AI-Powered Business Operations Platform

Desktop operations hub for managing events across Meetup, Eventbrite, and Headfirst Bristol — edit once, publish everywhere. 110+ MCP tools, AI augmentation, browser automation. 6,300+ automated tests. Entirely AI-directed development — I never touched the Node.js code directly.

SocialiseHub main interface showing event management dashboard

SocialiseHub analytics dashboard with revenue and attendance trends

              110+
              MCP tools via SSE — LLMs query databases, trigger automation, execute business ops
            

              109
              Test files covering stores, routes, validation, automation, analytics — 6,300+ assertions
            

              3
              Platforms synced — Meetup (GraphQL), Eventbrite (DOM automation), Headfirst (scraping)
            

Meetup GraphQL API

Hub

Ready

Eventbrite DOM Automation

Headfirst Browser Scraping

Analytics Timeline

Member Growth

Monthly Revenue (£)

Attendance Heatmap — Day × Time

45+

Fill Rate by Category

3,112Total Members

282Events Synced

£855Total Revenue

45%Avg Fill Rate

# 110+ tools available via SSE at localhost:3000/mcp/sse

# Click a tool to execute it

Rule: Never pass raw AI/LLM responses to the frontend.

Raw API Response

{
  "eventName": "Speed Friending",
  "Event_Date": null,
  "PRICE": "5",
  "desc": undefined,
  "attendee_count": "42",
  "venue": { "name": "The Lanes" }
}

Safe for Frontend

Community Data

3,112 members · Top: hiking (35 avg), speed friending (42 avg), board games (28 avg) · Friday 7pm peak

Market Analysis

Bristol outdoor events trending +40% · Kayaking, climbing, wild swimming rising · Competitor gap: weekday daytime events

Financial Data

Avg ticket £5.50 · Paid events 3x revenue · Venue costs £50-150 · Break-even at 15 attendees

Speed Friending Social Fri 7pm · The Lanes · 42 going · £5

Electron 40 React 19 Express 5 SQLite MCP Server TypeScript

Interactive CV

4-player multiplayer

Card Battle Game as a CV

A Hand of Fate-inspired card battle game where recruiters play objection cards and the CV counters with achievements. Procedural audio synthesis at runtime — zero audio files. Multiplayer via WebSocket with QR code entry. All vanilla JS, zero frameworks.

No Node.js Experience

Candidate lacks traditional backend experience

OBJECTION 6

♦

6,300+ Automated Tests

Vibecoded a full Electron app with production-grade test coverage

COUNTERED 9

christosgalaios.github.io/InteractiveCV

Vanilla JS Canvas API Web Audio API Socket.io Express 3D CSS Transforms

Play It

SocialiseApp

548 tests

Social Event Discovery Platform

Full-stack PWA with real-time chat, AI-matched micro-meets, gamification (XP + levels + 5 skill trees), community tribes, and Mango — the interactive kitten mascot you've already met on this page.

AI-researched brand palette from #E3705B base terracotta:

React 19 Supabase Express Framer Motion Zustand

Visit Live Site →

↓ Click "Join as Guest" to explore ↓

app.socialise.events

Same codebase. PWA-ready. Works everywhere.

Agentic Workflow System

4 codebases autonomous

AI Agent Orchestration Framework — Built from 5 Years of Studio Experience

I spent 5 years inside game studios. I know what every role does and why it exists — from how a technical lead protects a team from scope creep, to how QA and devs communicate without friction, to how a producer keeps a sprint on track. I encoded all of that institutional knowledge into an autonomous AI system that runs my codebases while I sleep.

★

CEO Agent Reads status logs, evaluates efficiency, writes business directives. When a task is needed, recommends which specialist to hire and writes their system prompt — Architect for planning, QA for validation, Test Writer for coverage. Operates at the business level only.

ceo-directives.md

⚙

Manager Agent Runs the sprint skill. Breaks directives into ≤15-line task specs. Dispatches the right specialist for each task — Architect first, then Dev, then Test Writer, then QA. Dispatches code-reviewer subagents after every commit. Never implements.

sprint.md / .manager-feedback.md

✎

Architect Plans before a single line is written. Reads the codebase, identifies files to change, documents patterns to follow. Produces the implementation blueprint the Dev agents execute against.

▾

Dev Agents Implement against the Architect's plan using Sonnet. Commit early and often. Check sprint.md before and after every commit. Use Haiku for file search, Sonnet for implementation.

●

Test Writer Writes tests after each implementation commit — unit, integration, and edge cases. Runs the full suite. Reports coverage gaps to Manager before QA is called in.

✓

QA Agent Validates via 110+ MCP tools — queries the database, triggers browser automation, captures screenshots, tests UI state. Files bug tickets directly into the sprint board. VERIFIED or back to Dev.

              5
              Specialist roles — CEO, Manager, Architect, Test Writer, QA — all autonomous, all coordinating via shared files
            

              24/7
              Night Shift mode — agents work continuously while I sleep, committing and pushing working code
            

              3-tier
              Model routing — Haiku for file search, Sonnet for implementation, Opus for strategy. Cost-optimised by design.
            

Sprint Skill — Full Role Pipeline

Built a custom sprint management skill that coordinates the full pipeline: Architect plans → Dev implements → Test Writer writes coverage → QA validates via MCP → Manager VERIFIED or sends feedback. All asynchronous, file-driven, no human in the loop. The sprint file is the single source of truth. Manager queues the next task the instant the current one is verified.

CEO as Hiring Manager

The CEO doesn't just write directives — it decides which role to spin up and writes the system prompt. "This task needs an Architect — here's the prompt." The CEO encodes domain knowledge directly into each agent's persona. A QA agent gets a prompt that references production bugs. An Architect gets context about the codebase patterns they must follow. The CEO controls quality through prompt engineering, not micromanagement.

Token Budget Engineering

Context windows are the scarcest resource in agentic systems. Task specs are capped at 15 lines. Subagents handle exploration to protect the main context. Model routing is strict: Haiku for grep/search, Sonnet for implementation, Opus for architecture decisions. Night Shift agents commit and hand off before context runs out — logging exactly where to resume so the next agent picks up seamlessly.

Learned-From-Mistakes System

Every real failure becomes a permanent cross-project rule: "always rebase before pushing", "workflow changes must land on the target branch", "never pass raw LLM responses to frontend", "never add auto-triggered API calls". The CEO encodes lessons after every sprint. The agents have inherited 5 years of operational knowledge — they don't repeat mistakes, they inherit the scar tissue.

Live Sprint Simulation

CEO claude-opus idle

Manager claude-opus idle

Architect claude-sonnet idle

Dev Agent 1 claude-sonnet idle

Dev Agent 2 claude-sonnet idle

Test Writer claude-sonnet idle

QA Agent claude-sonnet idle

ceo-directives.md

No active directives

sprint-plan.md

Awaiting Architect

.manager-feedback.md

No active tasks

cross-project-rules.md

0 rules

Backlog

In Progress

QA Review

Done

Sprint simulation ready. Click "Start Sprint" to watch agents work autonomously.

SocialiseHub — Live Preview

Speed Friending Social meetup synced

Board Games Night meetup synced

Comedy Open Mic eventbrite draft

Claude API MCP TypeScript Multi-Agent GitHub Actions Custom Skills Hooks Subagents Architect Agent Test Writer Agent

DevGuide

Built in 2 hours

Autonomous Content Pipeline

Multi-agent Python system that publishes SEO articles daily. Zero human intervention. Zero ongoing cost. 44 articles live.

This compass tracks your cursor. On devguide.co.uk it doubles as back-to-top. I think about UX details like this.

⏰ GitHub Actions 03:00 UTC daily

→

🔎 Discovery 3 niches

→

✍ Content 1,200+ words

→

✅ Validation Quality gate

→

🚀 Publish site/ + RSS

44 articles published. Zero human intervention. Zero ongoing cost.

Latest Published Articles

Comparison Cursor vs Windsurf 2025: Which AI Code Editor Wins? Review Claude Code Review: AI Terminal That Actually Ships Guide Best MCP Servers for Claude Code in 2025 Review Replit Agent Review: Can AI Build Your App? Comparison AI Coding Assistants Compared: GitHub Copilot vs Cursor vs Claude Guide Haiku vs Sonnet vs Opus: Which Claude Model to Use

Python GitHub Actions Next.js Zero Dependencies

Visit Live Site →

Socialise Website

Marketing Landing Page

Pure vanilla HTML/CSS/JS (~1,100 lines). Mouse-tracking spotlight effects, scroll reveal animations, parallax depth. Custom domain, no frameworks needed.

www.socialise.events

HTML CSS JavaScript GitHub Pages

Visit Live Site →

Edge AI Robot

Work in Progress

Physical Computing with LLMs

LLM-brained autonomous robot on Raspberry Pi 5 with Hailo 10H neural accelerator. Computer vision, object detection, autonomous navigation.

Robot chassis with Raspberry Pi and components on workbench

Edge AI Robot side view with tank treads and articulated arm

Python Raspberry Pi Hailo 10H Computer Vision

[ ENGINEERING DISCIPLINE ]

Every Commit Is Guarded. Every Push Is Rebased.

Git Workflow

git checkout -b fix/price-guard

git commit -m "fix: price null guard"

git fetch origin develop

git rebase origin/develop

Conflict in events.ts — resolve → lint

git push --force-with-lease origin fix/price-guard

gh pr create --base develop --title "fix: price guard"

CI passes → auto-merge to develop

develop → main merge → deploy to production

Learned rules: "Never bare git push" · "Rebase and merge only — no squash" · "GITHUB_TOKEN merges don't trigger workflows — use workflow_dispatch"

Pre-commit Hooks

⚙ Type-check .ts/.tsx edits

🔒 Block .env file edits

⚠ Lint check

👁 Self-reflect on merge

GitHub Actions CI/CD

Push

→

Lint

→

tsc

→

Test 0/6,300

→

Build

→

Deploy

Test Coverage

SocialiseHub

6,300+

SocialiseApp

548

DevGuide

0 total tests across all projects

Release Pipeline (Koffee Cup — Meta VR)

Dev

→

Shared

→

Review

→

Partner QA

→

Live

Dual-lane parallel development: LiveOps hotfixes and features ship independently

Night Shift Mode

When I say "night shift", the agents enter autonomous continuous operation. They never stop.

23:00 User: "night shift" → agents activate

23:12 Agent fixes price guard bug, commits, pushes

23:18 QA validates via MCP, marks VERIFIED

23:20 Manager writes next task, Agent picks it up

23:45 Backlog empty → Agent hunts for bugs proactively

00:10 Found missing validation on event dates → fixes, tests, commits

01:30 QA finds no more issues. Agents stay on standby, file watchers active

03:00 Context running low → commit, push, log next steps for next agent

08:00 User wakes up. Code shipped. Tests green. Status log updated.

Never stop working When done, find more work Commit early and often Log everything to status-log.md Only the user can stop night shift

[ EXPERIENCE ]

Where I Learned How Teams, Pipelines, and Products Actually Work

Feb 2026 – Present

AI Automation Engineer

Socialise Platform

Multi-agent orchestration across 4 codebases with autonomous sprint management
Production MCP server with 110+ tools enabling LLMs to execute business operations

Jan 2024 – Feb 2026

Technical Lead

Koffee Cup Ltd

Led 12 engineers (4 direct + sub-teams) in Meta VR ecosystem
Designed asset validation system eliminating manual re-export workflows
Primary technical contact for Meta — contributed to "Elite" vendor status
6-environment release pipeline with dual-lane parallel development

Jan 2023 – Jan 2024

Interactive Developer

Koffee Cup Ltd

Promoted to Technical Lead within 12 months
Haptic proximity system cited by client as key factor in prototype expansion to full product

Jan 2021 – Jan 2023

Software Engineer (Unity)

Virtually Sports

Shipped 2 commercial titles as sole engineer — full ownership of URP/HDRP pipelines, VR porting, and C++ backend integration
99.9% uptime on Unity video streaming architecture for high-traffic global betting platforms

[ EDUCATION ]

BSc (Hons) Computer Games Programming

First Class Honours — Middlesex University, London

IEEE Conference Publication (15 citations): "Developing an Educational Programming Game for Children with ADHD"

Key Project: Combat AI using Behaviour Trees and Finite State Machines

Mechanical Engineering of Automation

Undergraduate Coursework (2014–2016) — University of West Attica, Athens

Let's Build Something

christosgaleos@gmail.com LinkedIn GitHub