Zaniar Ahmadi · AI and cloud architect
AI agents that do real work. Built to a bank's standard.
I'm Zaniar. Inside a Tier-1 global bank I architect the platform where AI agents own whole workflows. The first one took vendor onboarding from five days to under an hour and removed 1,200 manual tasks a week. I build the same for businesses that can't hire a team like that.
Fifteen years shipping systems in banks, insurers and telecoms. Fixed scope and price before any build. If it isn't worth building, I'll say so.
Agent run
vendor-onboarding
Production AI inside a Tier-1 global bank
- Vendor onboarding, intake to completion
- 5 days → 1 hr
- Manual tasks removed every week
- 1,200
- Saved a year, audited and governed
- £400k
- IT support queries resolved with no human involved
- 40%
Experience includes
- Royal London Asset Management
- Lloyd's of London
- BP
- Vodafone
- Octopus Investments
What I build
Four things, in the order most businesses need them.
Agents that own a whole workflow.
Intake to completion, not a chatbot bolted on the side. Human review where it matters, an audit trail everywhere, and the repetitive parts done while your team sleeps.
Assistants that answer from your documents, with sources.
Policies, contracts, reports, tickets. Every answer cites where it came from and stays inside the data you allow, which is how it has to work with sensitive financial information.
The platform underneath.
Cloud landing zones, pipelines, monitoring and an evaluation harness, so the first agent is not the last and nothing breaks silently.
A straight answer on what's worth building.
Most AI projects fail at the scoping. I tell you what to build, what to fix first, and what to leave alone.
Work that shipped
The current client stays unnamed. The numbers are real.
Tier-1 global bank
Vendor onboarding: five days to under an hour.
An agent platform on Azure AI Foundry where each agent owns a business workflow end to end. The first agent took vendor onboarding from intake to completion, removed 1,200 manual tasks a week and saves around £400k a year. Governed and audited, with human review points built in.
Tier-1 global bank
5 days → 1 hr
1,200 manual tasks a week removed · around £400k a year saved
Tier-1 global bank
An IT helpdesk that answers itself.
A Copilot Studio assistant that resolves 40% of inbound IT support queries with no human involved, and shortens the queue for everything else.
Tier-1 global bank
40%
of inbound IT support queries resolved with no human involved
Asset manager
Research teams asking their data questions in plain English.
A retrieval agent that answers quant and research questions from structured and unstructured financial data, with sources, on a greenfield Azure data platform I built from scratch.
Asset manager
Plain English
over structured and unstructured financial data, every answer cited
Smaller businesses
The same discipline, scaled down.
Two consultants got their weekends back from automations that write the notes and send the follow‑ups. A kitchen showroom found out its agency's cost per customer didn't match the data. A security firm got advertising that made the phone ring and showed what each call cost.
Smaller businesses
Scaled down
two consultants, a kitchen showroom, a security firm
Tier-1 global bank
5 days → 1 hr
1,200 manual tasks a week removed · around £400k a year saved
How it works
Three steps. No deck.
Step 1 of 3
A scoping call about one workflow.
Thirty minutes, free. Bring the process that hurts: onboarding, checks, documents, a queue your team clears by hand. No preparation, no deck.
What we cover
- What happens today, step by step
- Who touches it and where it stalls
- What a good outcome looks like
Step 2 of 3
A straight answer.
Worth building, not worth it, or fix the data first. You leave with a decision and the rough shape of the build.
You leave with
- One decision: build, don't, or fix the data first
- The shape of the first agent
- What it plugs into
Step 3 of 3
Fixed scope and price, then I build it.
In production, not a demo: review points, an evaluation harness, monitoring, and a handover your own team can run.
What ships
- Agent in production with review points
- Evaluation harness and monitoring
- Handover your team can run
Zaniar Ahmadi · London
Why I do this
Banks pay for the discipline. Most businesses can't hire it.
I spent fifteen years making software ship safely in places where it can't fail: a telecoms email platform with a million users, payments, insurers, asset managers, and now a Tier-1 global bank, where I architect production AI.
AI is the same job with a new tool. The difference between an agent that works and a demo that gets switched off is the discipline underneath: access, testing, review points, monitoring. That is the gap I fill.
Experience includes Royal London Asset ManagementLloyd's of LondonBPVodafoneOctopus Investments
Before you call
Things people ask
Bring me the workflow your team dreads.
One free scoping call. A straight answer. Then a fixed scope and price if there's something worth building.