What I do
I am Alexey Shurov, an applied AI engineer based in London. I design, build, deploy and operate AI agents that run in production for paying clients, in the United Kingdom and internationally. The focus is not demos. It is systems that hold up under real volume and stay accurate enough to trust on a deadline.
- Production AI agents built end to end. Design, build, deploy, then operate and keep running.
- Document and email heavy workflows turned into automated pipelines. Reading, checking, reconciliation, pricing, order entry and reporting.
- Multi-agent systems with a person in the loop where the decision needs an owner.
- Integration with the systems a business already runs, including ERPs, MSSQL, Microsoft Business Central, Google Sheets, Outlook and messaging channels such as WhatsApp.
- A proof of concept hardened until it survives production, with evals, replay regression, retries and traceability to source.
Results, client confirmed
Read each figure as before to after. The work is measured by what it removes.
| Sector | The work | Before | After | What changed |
|---|---|---|---|---|
| Staffing | Weekly payroll run | 10 on the run | 2 | About 80 percent less manual work |
| Manufacturing | Quote turnaround | 1 to 3 days | Under 3 hours | About 90 percent faster, per quote |
| Wholesale distribution | Order entry | 10 | 1 | About 90 percent less manual work, over 95 percent line accuracy |
| Multi-country distribution | Order entry, per country | 5 | 2 | About 60 percent less manual work in every country |
| Global asset manager, over $1 trillion in assets | Specialist matching | About 3 days | About 1 hour | About 95 percent faster, a person owns the call |
| Field operations | Multi-site reporting | About 3 days | 10 minutes | About 99 percent faster, traceable to source |
Each system is written up under solutions, with how it works and what it needed from the client.
When to get in touch
- You need a production AI agent or a multi-agent system, not another prototype.
- A document or email heavy workflow is eating your team's week, such as timesheets, invoices, purchase orders, quotes, order entry or reporting.
- You have an AI proof of concept that works in a demo and needs to be reliable with real users.
- You want AI that reads documents, processes orders, prices from a catalogue, matches records or writes reports, with a person in the loop where it matters.
- You want an engineer who understands operations as well as models.
How I work
Every claim traces to a source, and releases ship behind replay tests on real past cases. In regulated work a person owns the decision. The agent recommends, with its reasoning cited.
Systems are built to stay up. A run can fail and recover without corrupting data. I do not claim a model never fails. I build so that when it does, it recovers safely.
I scope the first version to one workflow and one clear outcome so it ships in weeks rather than quarters, run it in shadow mode alongside the team, then widen it.
What a project needs from you
Access to the systems the work already lives in, usually a mailbox and the system of record, with a login the agent can use safely. The rules written down, or a person who knows them and has an afternoon to explain them.
A set of real past cases to test against. Replaying last month through a new version before it goes live is what lets a team trust the change. A named owner for the decisions the agent will not make, and a period where it runs alongside the team in shadow mode before it acts on anything.