Independent AI systems engineering · Tayyeb Mahmud, UKMove across the machine to see what keeps it running
I build AI systemsyour operations run on.
For COOs, CFOs, CTOs and heads of ops
I design, build and roll out AI platforms, agent workflows and enterprise integrations for finance and operations teams. On your data, inside your existing permissions, still running in daily use after I leave.
The machine is sealed. Move across it to see what keeps it running: permissions, checks, audit trail, monitoring.
Before anything is built: which workflow, what its errors cost, and whether your data and permissions can carry it. If the honest answer is don’t build it, that is the answer you get — before anyone has spent a budget.
Scoped, or stopped
02 / Architecture
Make the second use case cheap.
Models, data, tools, permissions and your own definitions become one operating layer. The first use case pays for that layer; the second and third inherit it instead of starting again.
Platform in place
03 / Controls
Agent proposes, verifier gates.
The model drafts; deterministic checks decide what proceeds. Anything that fails stops and routes to a person with the reason attached, logged and reversible — because finance work has no acceptable error rate.
No unchecked writes
04 / Handover
Make it run without me.
Real users, production data, monitored runs, your team building alongside me. One person is a limit as well as a feature: everything is documented and owned by your team, so it does not depend on me being available.
Yours to run
01.1
Business pressure
The outcome worth changing.
01.2
High-value workflow
One bounded place to begin.
01.3
Risk boundary
Consequence made explicit.
02.1
Data + context
Governed inputs with lineage.
02.2
Models + tools
Routed through one reusable layer.
02.3
Permissions
Every capability is bounded.
03.1
Deterministic checks
Known rules before model judgment.
03.2
Human gate
Review where consequence demands it.
03.3
Audit + rollback
Traceable, repeatable, reversible.
04.1
Monitoring
Behaviour stays observable.
04.2
Escalation
Exceptions reach the right owner.
04.3
Team handover
Your team can operate and extend it.
01.1Business pressureThe outcome worth changing.
01.2High-value workflowOne bounded place to begin.
01.3Risk boundaryConsequence made explicit.
SELECTED SYSTEMS / 05—08
What was built, and what it changed.
One sync change forecast to save £2–4M a year in token spend, reviewed and accepted by a member of the client’s finance team. Finance answers that land in the meeting. Access for 1,000+ people driven from the HR record. A frontend brief that became an AI product line.
D-0105 / 08
AI, data and automation enablement across a global agency
1AYM contracts into a PE-backed global media and marketing agency, working alongside its finance, systems, data warehouse and internal AI tooling teams. The AI programme is the agency's own — run by their leadership, with skills authored across the business. My contribution is the engineering underneath it: finance systems and data warehouse enablement, identity provisioning, verifier-gated automation for finance-critical tasks, and platform optimisation where I found it — including replacing a per-user API call pattern with a single ten-minute sync, forecast to save £2–4 million a year on token consumption, reviewed and accepted by a member of the client's finance team. The internal skill platform I contribute to now carries around 600 production skills across business-unit plugins.
Self-serve finance answers, without breaking permissions
A CFO wanted the wider business to get answers about client and financial performance without queuing for a member of the finance team. I built a connector between the company's AI workspace and its existing Looker setup that inherited Looker's permission model exactly — people with privileged access kept it, everyone else saw only what they were already entitled to see. Questions that previously took around two hours to come back from finance now resolve in the meeting where the question is asked. Connecting the data took an afternoon; the week that followed was spent building the semantic layer that made the answers trustworthy.
~2 hrs → in the meetingTime to a governed finance answer
Identity provisioning for 1,000+ users across 600+ groups
Access to Notion and Claude across the organisation is provisioned automatically from the HR system, covering more than 1,000 users and over 600 groups, where a single person can belong to many groups at once. The agency's internal IT team was over capacity and could not take the build on, so 1AYM built the system from scratch rather than forcing the model into off-the-shelf group tooling. It has run reliably for seven months. Doing this membership matrix by hand was never realistic — and manual provisioning is how leavers keep their access.
A frontend contract that became an AI product line
I was contracted to build the frontend for a government-backed education company's website in the Middle East. Working inside the product, the more valuable opportunity was obvious: their mock test product could do more than mark answers. I built an AI tutor that speaks the languages their learners already speak and teaches English in the learner's own language, rather than assuming English to teach English. That turned a defined frontend piece into a long-term engagement — I now work directly with the CEO and the team on where AI belongs across the business, and the capability has given them something concrete to show in investor and grant conversations.
Frontend briefHow the engagement started
MultilingualTutor teaches English in the learner's first language
Connecting AI to your data now takes an afternoon. That ease is the trap. Everything that decides whether the answers are usable — which table is the real revenue, who may see it, what happens when it runs a thousand times a day — is still sitting in people’s heads.
No logo wall, no testimonial you cannot trace. Two proctored Anthropic exams — Architect (Professional) and Associate (Foundations), both issued July 2026, both open on Credly. Client names are held back by agreement, not for want of references — bring them up on the call and I will tell you what I can.
Thirty minutes with the person who would build it — no deck. Bring one process and what a silent error costs you. You leave knowing whether there is a system worth building, what it would have to survive, and whether I am right for it.
