Meet Meliah, my personal AI assistant. She reads the mail and chats, briefs me, drafts replies and answers out loud. Under her: an HRIS with payroll, an ERP with its own POS and a workflow and ticketing platform, all built in-house with AI agents as my dev team.
Meliah's real voice, made on her own PC (Kokoro) · questions heard on-device (Whisper) · sample data
She runs on my own PC, with a phone companion. She works through my mail, chats and sales data around the clock, then briefs me, drafts for me and answers by voice. Built with Claude Code and Python. Her voice never leaves the machine.
Outlook, Teams, WhatsApp, Viber and years of old Lark mail flow into one linked, searchable knowledge base, with a hub for every person, branch, vendor and topic. Passwords and codes are redacted on the way in.
At 7:30 she lists the top three, who's waiting on me and today's meetings, each with one-click Reply · Task · Done. Before each meeting there's a prep brief with the open threads and last decisions.
Tasks come from mail, chat and meetings and keep a dated history. When someone writes "I'll send it Friday", she proposes a task; it's only added when I tap Add.
A real back-and-forth: about 3 seconds from the end of a question to her first spoken word. Whisper hears and Kokoro speaks, both on my own PC. Private items are never read out loud.
Robots sign in to each POS back-office, pull the daily sales and product mix, and load one sales database she can answer from. Missing days stay missing, never zero. The same crawling engine (headless browser, saved sessions, retries, checks) takes on any other portal or report that has no API.
A phone app shows a private copy of the brief and tasks. Questions spoken on the road are answered by the PC and come back in her voice. The app itself holds no data.
A replica of the desktop app with made-up people and numbers. Nothing connects to a server.
Agents and robots do the gathering, sorting, chasing and first drafts. Anything that leaves the building, moves money or touches a POS waits for a person.
Mail, Teams, WhatsApp and Viber are filed under the right person, branch or vendor, with secrets redacted.
Sign in, export each branch's sales and product mix, re-check recent days for late uploads, and flag gaps.
Top three, overdue, waiting-on, today's meetings, on the PC and the phone.
Who's in it, the open threads, and what was decided last time.
Commitments in messages become proposed tasks, so nothing promised slips.
Ask on the road; the answer comes back in her voice, made on the PC.
What moved, what slipped, and tomorrow's first move.
She reads the day's questions and updates her answer playbook. When the data and the playbook disagree, the data wins.
Geofenced, selfie-checked punches roll into semi-monthly payroll with the government tables; buddy-punching gets flagged.
Each step finds its approver and arrives as a Teams card. Nobody approves their own request.
Supplier price changes are flagged before they reach the margin.
The POS robots' engine is reused for other portals, exports and checks: headless browser, saved sessions, retries.
The HR and operations system, built from the ground up and used across every branch. It's now becoming MLG Hub, one front door for every company app.
Best on a computer; on a phone, open it full screen.
The same approach for approvals, help desk, purchasing and selling: each one replaced spreadsheets, chat threads or a tool that didn't fit.
One place to ask for anything: an IT or HR ticket, a payment request, a job offer. Each request finds its own approvers and reaches them where they already work.
An ERP laid out like SAP (FI, CO, MM, SD and the rest) across three companies, with its own point of sale and a self-order kiosk.
Replicas with sample data. Best on a computer; on a phone, open them full screen.
I design the systems and make the calls. Claude Code sessions do most of the typing, each one owning one codebase. Everything they write goes through tests, checks and my review before anyone at work sees it.
I write the decision and the reason into the system's single source of truth.
The agent that owns that codebase plans the change, writes the code and its tests.
Tests, type checks, scans for secrets and real data, screenshots at phone and desktop widths.
I read the diff. Database migrations I run on production myself.
A dated change-log line and a handoff note, so the next session starts where this one stopped.
A representative session, condensed. The code stays private; the process is real.
Who sees what is enforced with row-level security and grants, never by hiding a button.
Integer centavos or exact decimals, and tax rules built into the schema.
Sales, approvals, notes and audit logs gain new rows; they're never edited away.
Agents draft and prepare. People send, sign and pay.
Before touching another system, an agent asks its owner for the history; only my decisions get relayed.
Architecture, decisions with reasons, and a dated change log the agents read first.
IT Head & Digital Transformation Tech Head
I lead IT and digital transformation for a multi-brand company with 35+ branches, and I design, build and roll out its own platforms, replacing legacy and third-party tools with systems built for how we actually operate.
I build AI into that work: an assistant that runs my day, automations that do the routine, and AI agents as a development team. The same patterns fit any organisation with people, money, approvals and many locations.
Leads IT operations, the in-house platform suite and AI across every brand and branch.
Architected and built the HRIS from the ground up; added approval workflows, IT & HR ticketing and asset management; owned rollout, change management and the guides that drive adoption.
IT infrastructure, point-of-sale operations and support for 35+ branches; led the IT team and vendors; drove the company-wide move to Microsoft 365.
Started building the company's own systems.
Joined the group's IT team.
Contact
The real code runs a business, so it stays private, but I'm happy to walk you through any of it, Meliah included.