Tao Huie-Miles

IT support and infrastructure engineer · escalations, Microsoft 365, networks and automation

Maidstone, Kent · [email protected] · seyta.ukseyta.uk

Enhanced DBS held: can start with no clearance delay

Open to hybrid roles in Kent and London

370+client businesses supported
400+devices refreshed across two school sites
0ports open to the internet on my AI platform
4,500+automated checks guarding my AI platform

Profile

I'm the engineer critical outages get escalated to. At an MSP supporting 370+ client businesses I look after Windows servers, firewalls and switching, Microsoft 365 and Entra ID, on site and remote, and take failures through to the fix.

Before that I ran a school campus network, led a 400+ device refresh across two sites and moved the school from shared drives to SharePoint Online and Teams. I automate repeat work with PowerShell and Power Automate.

Outside work I designed and run The Foundry, my own AI platform: zero open ports, failover to Azure, 3,000+ automated checks, and a team of AI agents I direct to build it. I do my best work in the flow: give me a problem and the room to run with it, and I stay on it until it's solved. I'm looking for an infrastructure role where I own one environment end to end and automate the repetitive work away.

Away from work: music of every kind (hyperpop to jazz-funk), Valve games, and building the Foundry live on Twitch.

Key skills

Servers & identity: Windows Server · Active Directory · Group Policy · Entra ID (hybrid join) · DNS · DHCP · RADIUS · Hyper-V
Microsoft 365 & cloud: Exchange Online · SharePoint Online · Teams · Intune · Windows Autopilot · Azure · Google Workspace
Networking & voice: SonicWall · FortiGate · Sophos · Cisco · HP/Aruba · Ubiquiti UniFi · VLANs · VPNs · 3CX · Teams Phone
Security & mail: Defender for Business · Huntress · Sophos Central · SPF, DKIM and DMARC · Smoothwall and Lightspeed filtering
Automation & monitoring: PowerShell · Power Automate · Python · n8n · OpenTelemetry · Azure Application Insights · Power BI
Backup, endpoints & service: Veeam · Synology and QNAP NAS · Mosyle · Jamf · Apple School Manager · PaperCut · ConnectWise · Freshdesk · ITIL SLA management
Platform & AI: Cloudflare Tunnel and Access · OAuth 2.0 / OpenID Connect · FastAPI · Azure AI Foundry · Model Context Protocol · local LLMs (LM Studio) · multi-agent orchestration (Claude Code, n8n)

Experience

Technical Support Engineer · Akita IT

Feb 2025 – present · Maidstone, Kent

Escalation point for 370+ client businesses, on site and remote: the tickets first line can't close and the outages that can't wait. Part of the team named SDI MSP Service Desk of the Year (SME).

Own the ticket. Second-line incidents and requests from first contact to the fix, worked by impact against SLAs, and the client kept told in plain English until they say it's sorted.
Own the outages. Critical server, network and cloud failures escalate to me, and I take them through to the fix.
Microsoft 365 and Entra ID, day to day. Mailboxes, licences, Teams and SharePoint permissions, 2FA lockouts and hybrid-join faults: the tenant work behind most of the queue.
Devices and identity. Intune and Windows Autopilot enrolment faults, RADIUS failures and broken profiles fixed with targeted registry edits, profile repairs and broker account recovery.
Automate the boring bits. PowerShell scripts and Power Automate flows for onboarding and offboarding, tenant admin, reporting and repeat config changes across Exchange, Microsoft 365 and Entra ID.
Write it down. Fixes written up as they're found, so first line closes the repeat next time and the client gets the same answer whoever picks up.
Invisible maintenance. Out-of-hours patching of Windows servers, backup systems and critical kit, so nobody notices it happened.
Networks top to bottom. SonicWall, FortiGate and Sophos firewalls; Aruba/HP, Cisco and UniFi switching; VPNs. Sometimes the fix is band steering on an access point; sometimes it's phoning the ISP and waiting politely.
First responder on security. Triage detections in Huntress, Defender for Business and Sophos Central: isolate affected machines fast, escalate the serious ones, and pick the quickest safe fix over the textbook one.
Mail that arrives. Deliverability across Microsoft 365, Google Workspace and third-party hosts, with SPF, DKIM and DMARC put right whenever mail lands in spam.
On site, hands on. Hardware provisioning, rack and patch work, NAS installs and enterprise print deployments (PaperCut, Universal Print).
Into the cloud. Help clients deploy and manage their resources in Azure (Azure Resource Manager) as their workloads move to the cloud.
Plan ahead. Advise on the fix today and plan for possible scoping upgrades and equipment changes before they turn into outages.

IT Network Technician · Tunbridge Wells Grammar School

Feb 2023 – Jan 2025 · Tunbridge Wells, Kent

The whole site's support, in house. First call for every staff member and classroom: desktops, laptops, iPads, interactive panels, printing, Wi-Fi and accounts, fixed at the desk or from the office.
Led a two-site hardware refresh. 400+ Windows desktops and laptops, interactive classroom touchscreens and iPads, planned and rolled out in stages.
Ran the campus network. Cisco and HP/Aruba switching, with separate VLANs for staff, student BYOD, VoIP and CCTV.
Moved the school off shared drives. Migrated to SharePoint Online and Teams, designed the new security groups and policies, then trained staff in person and stayed for the awkward questions.
400+ smart devices, many tools. Macs, iPads, iPhones and BenQ interactive panels, each family with its own management tool: Mosyle, briefly Jamf, and Apple School Manager for over-the-air profiles and app licensing.
Identity and recovery. Active Directory hybrid-joined to Entra ID, Group Policy, and disaster recovery when it was needed.
Physical security. Axxon CCTV and security sensors on isolated camera networks, with video storage kept within retention schedules.
Safeguarding online. KCSIE-compliant web filtering (Smoothwall, Lightspeed), with SSL inspection certificates deployed to every workstation.
The unglamorous essentials. Cat6 terminated and tested, patch panels and UPS systems maintained, storage health and backups always watched. Also trained staff on Bromcom, the school's new MIS.

Technical Support / Logistics Technician · Optomany (DNA Payments)

2022 – 2023 · Kings Hill, Kent

Kept shops taking payments. Support for EPOS hardware and payment terminals (Pax, Verifone, Ingenico).
Network detective work. Diagnosed payment faults across merchant LAN/WAN: DHCP conflicts, gateway dropouts, packet loss.
Ran the queue. High-impact retail outages first, replacement stock coordinated, gateway partners kept in the loop.

Project

The Foundry · my home AI platform

Live at seyta.uk ↗Live at seyta.uk

A multi-service AI platform I designed and run on my own hardware, with friends using it through a public web chat and Discord.

Zero open ports. Every service listens on 127.0.0.1 only. An outbound Cloudflare Tunnel is the one way in, with Cloudflare Access at the edge. Each private service then checks identity again itself: the Portal verifies Cloudflare's signed token, the gateway its own API key.
Access control. One gateway in front of the model enforces API keys, a rate limit of 120 requests a minute, and route and host allowlists. Friends sign in with Discord or Google (OAuth 2.0, OpenID Connect) and wait in an approval queue, with per-person limits and a daily cap on cloud replies.
Failover. A local model on one RTX 3090 answers first. When the GPU is busy or the model stalls, gpt-5-mini on Azure AI Foundry carries on the same conversation.
Monitoring. An owner dashboard with live service status, GPU memory history, p50 and p95 answer times and fallback rate, plus OpenTelemetry traces in Azure Application Insights. Telemetry never stores message text.
Operations. Nine services started, stopped and health-checked by idempotent PowerShell scripts, each managed by its recorded PID, with written reboot, key-rotation and troubleshooting procedures.
Change control. 3,000+ automated checks, attack tests included, gate every change. Every commit is scanned for secrets first.
Multi-agent orchestration. Describe a job and an Architect designs the flow (which models, in what order, inside my cost limits); n8n runs it, and a headless browser opens anything it builds and reads its console before it counts as done.
Measured, not guessed. My own benchmark found a localhost/IPv6 delay and cut the wait for the first word from 4.15 s to 0.16 s.
Built by directing AI agents. Claude, Gemini, ChatGPT and local models share a ledger and a task board, hand work to each other mid-task, and follow my rules on what needs my OK.
Stack. Python, FastAPI, SQLite, Git, PowerShell, n8n, LM Studio, Azure AI Foundry, and a Model Context Protocol server with 16 tools.