I build LLM systems in Python and I measure whether they work. Chatbots that answer from your own documents: parsing, chunking, embeddings, Qdrant, and an answer that cites the page it came from. Multi-agent pipelines on LangGraph with tool calling and structured output. Scrapers on Playwright that survive Cloudflare. I also build automation that runs with nobody watching it. One of mine works nine platforms at once: it finds target communities, verifies them, posts on a quota it measured itself, reads every reply, answers the first message and files people into a database — 9 194 communities screened down to 2 822 live targets, and a report twice a day instead of an operator. My own system runs in production at ortant.net with open code on GitHub. Its quality is a number, not an adjective: a 379-example gold set, macro-F1 0.81, a gate that stopped 7 runs out of 31, median run cost $0.13. Price and deadline up front. They do not grow later.…

Aysen Khovrov

I build LLM systems in Python and I measure whether they work. Chatbots that answer from your own documents: parsing, chunking, embeddings, Qdrant, and an answer that cites the page it came from. Multi-agent pipelines on LangGraph with tool calling and structured output. Scrapers on Playwright that survive Cloudflare. I also build automation that runs with nobody watching it. One of mine works nine platforms at once: it finds target communities, verifies them, posts on a quota it measured itself, reads every reply, answers the first message and files people into a database — 9 194 communities screened down to 2 822 live targets, and a report twice a day instead of an operator. My own system runs in production at ortant.net with open code on GitHub. Its quality is a number, not an adjective: a 379-example gold set, macro-F1 0.81, a gate that stopped 7 runs out of 31, median run cost $0.13. Price and deadline up front. They do not grow later.…

Available to hire

I build LLM systems in Python and I measure whether they work.

Chatbots that answer from your own documents: parsing, chunking, embeddings, Qdrant, and an answer that cites the page it came from. Multi-agent pipelines on LangGraph with tool calling and structured output. Scrapers on Playwright that survive Cloudflare.

I also build automation that runs with nobody watching it. One of mine works nine platforms at once: it finds target communities, verifies them, posts on a quota it measured itself, reads every reply, answers the first message and files people into a database — 9 194 communities screened down to 2 822 live targets, and a report twice a day instead of an operator.

My own system runs in production at ortant.net with open code on GitHub. Its quality is a number, not an adjective: a 379-example gold set, macro-F1 0.81, a gate that stopped 7 runs out of 31, median run cost $0.13.

Price and deadline up front. They do not grow later.

See more

Work Experience

Add your work experience history here.

Education

Add your educational history here.

Qualifications

Add your qualifications or awards here.
    Data extraction from sites behind anti-bot protection

    Stable collection from sources that actively resist scraping. Retries, session handling and rate control, so a blocked page becomes a logged gap rather than a silent hole in the data.

    Coverage is proven, not assumed: every run reports what was collected and what was missed, so the client sees the shape of the data instead of a number that looks complete.

    Stack: Python, Playwright, PostgreSQL, Docker.

    Measured output quality for an LLM pipeline

    A publication gate for generated content. The pipeline produces text, an automated judge scores it, and anything below the bar never reaches the client.

    Quality is a number here, not an opinion: a 379-example reference dataset, macro-F1 0.81, LLM-as-a-judge scoring. The gate stopped 7 runs out of 31 before anything was published. Median run cost $0.13 and took 12.8 minutes.

    Stack: Python, LangChain, OpenRouter model routing, PostgreSQL.

    Retrieval-augmented search over 10,000+ documents

    A question-answering system over a large document archive. Document parsing, chunking, embeddings and vector search in Qdrant, with every answer pointing back to the exact source page so the client can verify it instead of trusting it.

Hire a AI Engineer

We have the best ai engineer experts on Twine. Hire a ai engineer in Almaty today.