For business
The most common questions from owners and managers thinking about AI.
Where do we even start if we've never tried AI?
With a discovery session — it's free, 30 minutes. I listen to your processes and tell you honestly: is there anything here for AI at all, or is it still too early. Often the honest advice is to first put your CRM and processes in order, and only then think about automation.
Do you do the implementation or only advise?
With me you build the strategy: where AI gives leverage, what to do first, the plan for 6–12 months. The implementation itself is run by my brand teams — Auspex (CRM and processes) and Grow2.ai (AI agents). A ready brief is handed over, so you won't have to explain everything from scratch.
How much does it cost?
Advisory has three tiers: a one-off consultation $300–500, an AI strategy $1.5–3K (project), an advisory retainer $2–4K/mo. I quote the exact figure after a short brief. The cost of implementation is estimated separately by the Auspex/Grow2 teams — for the specific task.
How long does it take?
A consultation — one 90-minute call plus prep. A strategy — 2–3 working sessions over a few weeks, with a document as the output. Implementation timelines depend on the task — the relevant brand team estimates them honestly after discovery.
Who owns the results of the work?
You do. The strategy document, the priority map, the implementation brief — they're yours. In implementation projects it's the same: the code, prompts and architecture are handed over to the client. I don't want to be «the only person who knows how to fix it».
And what about data confidentiality?
An NDA — no problem, on your template or mine. In projects the client's data is not used to train models, and the architecture is built around your storage requirements — down to on-premise, if it's really needed.
About courses
What's happening with training and when the course will be ready.
Do you have courses?
Not yet — and I don't pretend I do. The first course is in the works: hands-on Claude Code, from everyday tasks to your own agents. Leave your email on the courses page — I'll let you know about the launch first.
What to do while there is no course?
Read the blog and cases — that's real practice with numbers and mistakes. Subscribe to the newsletter: 2–3 emails a month, specific prompts and honest thoughts. From time to time I run webinars — announcements will be in the newsletter.
Technical
Stack, integrations, limitations. For those who know their way around.
What stack do you use?
The main LLM is Claude (Sonnet 4.6), orchestration is Claude Code with custom skills. RAG — vector databases plus knowledge graphs for agents' long-term memory. Integrations — via API and n8n. It's all pragmatic: if you already have a different stack — I'll adapt.
Can it run on an open-source LLM (Llama, Mistral)?
Yes, if there are good reasons: critical confidentiality, very high volumes, offline mode. In ordinary cases Claude comes out cheaper and better. But I'm not religious about it — I choose what's better for the specific task.
How do you fight hallucinations?
Three levels: 1) a supervisor agent (LLM-as-judge) checks answers against the knowledge base; 2) the «don't know — hand off to a human» policy is written into the system prompt; 3) regression tests on real dialogues, run on every prompt update. This isn't magic, it's engineering discipline.
Do you build CRM integrations — Bitrix24, amoCRM, Salesforce?
CRM is home turf: Bitrix24 (cloud and on-premise), Uspacy, amoCRM, Monday.com, Salesforce — that's the Auspex profile, with over a thousand such projects over the years. Exotic systems — I'll show an honest estimate at discovery.
Didn't find an answer?
Write to hello@amaryasov.expert or on Telegram — I'll reply.