I design and build LLM-based AI systems to automate business processes and help teams apply them in ways that create real business value.
System ingests a company list and automatically identifies relevant decision-makers: aggregates public and specialized sources, normalizes roles and contacts, runs initial verification, and exports results in a sales-friendly format.
Goal was to validate whether semantic search would improve quality in a highly specialized domain. The assessment showed that baseline embeddings deliver weak relevance here; acceptable quality would require domain adaptation / a larger model plus additional data work—not cost-effective at this stage. Recommendation: don't implement semantic search.
Goal was to quickly understand competitors' technology stacks and solution architecture. Built an AI agent system that gathers evidence across sources, structures insights, and produces a final report with a component map and architecture diagrams.
Automation finds the right organization listing using only name and address, pulls ratings/reviews, and writes them into an internal database. The hard part is reliably matching the correct listing, so the flow includes fuzzy name/address matching plus multi-step validation to handle edge cases: multiple businesses in one building, renames, franchises, and address mismatches.
Automated contract checks against a checklist, key-field extraction, and structured export for further work.
A Telegram bot for photo studios: staff sends a client photo, the system automatically prepares it to passport standards—cropping, lighting, and background adjustments.
An AI agent monitors selected news sources daily and delivers a structured morning brief of the most important updates in a convenient format.
I work primarily with LLM-based systems and generative AI — from ChatGPT-level tools to custom workflows and automations. My focus is on tasks and processes that can realistically be improved with AI, not on "AI for the sake of AI".
I help you understand where LLMs can support your work — and where they can't.
This is a consultation with preparation: I review your context and cases in advance, so our discussion is focused on decisions, not explanations.
A good starting point for an AI task audit — whether for business processes or personal workflows.
A deeper analysis of selected AI or LLM use cases, without building a production system.
The goal here is to think through a case properly and explain how it may work in practice — including constraints, risks, and trade-offs.
A good fit if a single consultation isn't enough and you want clarity before deciding whether to build.
Hands-on validation of a selected AI or LLM use case through a working prototype or automation flow.
At this stage, we move from reasoning to practice and see how things actually perform with real or semi-real inputs.
The deliverable can be a working prototype, a demo automation, or a solid understanding that the idea isn't worth scaling further. Next steps (full development, long-term support, or maintenance) are discussed separately and depend on the case.
I work with low-code platforms. This approach lets me focus on the product and business problem — not technical gymnastics — and ship faster than traditional development. The stack is always chosen based on the task, not the other way around.