Looking for a job

Евгений Ивентьев

@0008171

Message

Backend / LLM Engineer at Университет ИТМО; raised RAG fact recall from 65% to 79.9%.

Kyrgyzstan

Message

What I'm looking for

I’m open to remote roles and contracts where I can own the path from requirements and a quick prototype to production, especially for AI assistants and LLM systems.

At Университет ИТМО, I work on an agentic LLM platform for a major oil and gas company, where employees use AI agents and self-hosted models to search a knowledge base of roughly 3 million document chunks. I designed the unified agent runtime and fixed a race condition that was losing tasks.

I rewrote document parsing and built an indexing evaluation framework with statistical testing. Fact recall increased from 65.0% to 79.9%, with 33% more extracted text and no regressions; 25,924 documents were re-indexed in production.

At START LLC, I designed and built a fulfillment service connecting a warehouse, Wildberries, and 1C. I also developed marketplace data pipelines to support assortment and advertising decisions.

At Soft IT, I worked on data engineering and document automation for a container logistics group. My projects included invoice processing that cut processing time from 10 business days to 8 hours, an OCR platform, and an LLM document assistant that reduced lookup time from 15 minutes to 1–2 minutes.

Experience

Work history, roles, and key accomplishments

Current

Backend / LLM Engineer

Current

Университет ИТМО

Dec 2025 - Present (10 months)

Agentic LLM platform for a major oil and gas company: employees work with the knowledge base and documents through AI agents on self-hosted models, no external cloud APIs. RAG over ~3M document chunks. Team of ~25.

- Agent SDK: designed the unified runtime all platform agents run on (gRPC streaming, long-running tasks via RabbitMQ with cancellation and TTL, OpenAI-compatible API). Found and fixed

Python Backend Developer

START LLC

Feb 2025 - Nov 2025 (9 months)

Fulfillment and analytics platform for a Wildberries marketplace seller: 14 seller accounts, warehouse, 1C integration. Team of 5-7.

- Fulfillment service "warehouse -> Wildberries -> 1C": designed and built from scratch - assembly tasks, 4 supply types, QR labels, reservations, shipments to 1C. The warehouse runs all 14 accounts through a single service.
- Marketplace data pipeline: scheduled in

Python / Data Engineer

Soft IT

Oct 2021 - Jan 2025 (3 years 3 months)

Projects for a large container logistics group (terminals, rail): corporate data warehouse (parsers -> Apache NiFi -> ClickHouse -> DataLens) plus LLM and OCR solutions for document workflows.

- Invoice & archive processing: automated parsing of archives and Excel invoices with counterparty unification and loading into ClickHouse. Processing time dropped from 10 business days to 8 hours.
- Stream