Bench+
← Back to talent
M

BENCH+ PROFILE

Matvey

AI Integration Engineer
Production LLM pipelines, Text-to-SQL, RAG-based data access, model routing, and backend integration for analytics systems

Senior AI integration profile with hands-on production delivery of LLM-driven analytics pipelines, strong backend integration depth, and practical experience around RAG, model routing, evaluation, data access controls, and monitoring.

BEST-FIT USE CASE

Best suited for building or extending production LLM features that connect natural-language interfaces with enterprise data warehouses, APIs, validation pipelines, and operational monitoring.

Not the primary fit: Less aligned with roles focused primarily on model research, computer vision, mobile development, product design, or standalone frontend delivery.

EVIDENCE HIGHLIGHTS

What supports the assessment

01

Developed a production natural-language-to-SQL pipeline for cross-channel analytics over a large set of marketing data inputs.

02

Built architecture using MCP-style context access, dynamic RAG, and a vector database to support controlled interaction with warehouse schemas.

03

Implemented model routing so complex tasks could use stronger hosted models while common SQL-generation and parsing tasks could use smaller models.

04

Configured masking before proprietary model calls and automated release testing using judge-based evaluation.

05

Optimized relational database transactions and indexes and added in-memory caching for logistics-related backend workflows.

06

Built internal real-time operational dashboards using a SPA frontend, typed JavaScript stack, and WebSockets.

07

Supported backend stability through test coverage, code review, alerting, and monitoring setup.

WORKING STYLE

How they show up

  • Described an engineering cycle that starts with analyzing where an agent fails, then designing changes, adding checks or routing logic, testing against masked real cases, reviewing accuracy and latency, manually checking results, and deploying.
  • Explained LLM feature evaluation through a small representative dataset, baseline generation, token consumption review, smaller-model assessment, routing, and automated quality evaluation.
  • When evaluating weak implementation directions, described clarifying the business need, presenting trade-offs, preparing a small proof when useful, and proposing an alternative implementation path.