Artem Prianikov · AI Solutions Architect & Delivery Lead Get in touch

AI integration & development

A person spends the day as the glue between a dozen interfaces.

Those systems were never designed to talk to each other, so a human was put in the gap. Most of what I build comes down to taking the person out of it.

A working day, as most companies actually run it

Where the work arrives

Mail
Telegram
WhatsApp
Zoom
Calendar
Tasks
The human, doing the carrying moving numbers across, retyping what is already written somewhere else

Where it has to end up

CRM
ERP
Spreadsheets
Docs
PDF
Banking

The cause is structural, not a matter of discipline. Take the carrying away and the same team does more, with fewer mistakes and no new headcount.

How I work

A project, from the first meeting to hand-off

Full cycle either way, whether it is internal tooling or a system delivered to a client.

01

Diagnostic

A meeting with the owner or the lead, the process examined as it actually runs, then findings and the options that follow from them.

02

Proposal with prices

The client picks the scope; the price answers the scope they picked. Anything doubtful is called doubtful before the work starts, not halfway through.

03

Spec and delivery

The architecture is written down before anyone is assigned, so each task is doable as written. The part with the most risk in it I take myself.

04

Hand-off and training

A system nobody on the client's side understands is dead in six months. Hand-off is part of the job, not a courtesy after it.

Ordering it this way means the price answers a scope the client chose — which is what makes it defensible.

Background

Building automation into businesses since before ChatGPT existed

2019first work with AI, well before the tooling was mainstream
15projects taken through to client hand-off
3–4developers led: scoping, code review, mentoring
5+years of Python automation, 2+ building on LLMs

Selected work

What this looks like in practice

In production

AI front desk for a rental business

More than 50 apartments. The assistant handles guest conversations end to end, from the first question to check-in, and hands over the moment it is unsure — a confident answer from the wrong section is the worst failure there is.

~50 guest conversations a day

Retrieval at scale

A knowledge base over a catalogue of thousands of pages

Off-the-shelf tools start degrading somewhere around 200 to 300 pages. The fix was routing rather than retrieval: split the catalogue into semantic blocks and put an orchestrator in front that decides which block a question belongs to.

Thousands of pages, answered reliably

Internal tooling

A pipeline from sales call to scoped proposal

The hard part was never generation. A model will happily propose an architecture that cannot be built, or an iceberg whose hidden half doubles the cost. The work is keeping the result buildable by a specific team, priced for what it really takes.

~20 proposals a month · a full day cut to about two hours

Delivered solo

bannamtalay.com

A resort website, the CRM behind it and the AI assistant on top, designed and built end to end by one person. Live and earning its keep.

2–3 inbound enquiries a day

Tools

Pythonn8nLLM APIsClaude Code · Codex RAG pipelinesTelegram Bot APIWeb scraping LinuxDocker

Contact

Tell me what the day looks like now, and I will tell you what can come off it.

Thailand, GMT+7 · remote, open to relocation Russian (native) · English (professional working proficiency)