Dmitriy Mgar

Senior Fullstack Engineer / AI Engineer

8 years in data, ML & AI. MIPT graduate. Built AiKarta — a full-cycle B2C real estate platform with an AI agent — solo in 6 months using Cursor IDE and Claude. T-shaped engineer with deep expertise in Python backend, AI/ML, and solid frontend skills.

About

MIPT (Applied Mathematics and Physics, Master's 2017). ICAgile Certified Professional (ICP), 2021. Russian (native), English (B2).

Experience

  • AiKarta
    2025–present

    Co-founder, CTO / Solo Fullstack Developer

    Solo engineered a full-cycle B2C platform from DB architecture to production using Cursor IDE + Claude. SPA in Vanilla JS + Yandex Maps API (~100K objects). Async Python/FastAPI backend with a LangGraph agent. MCP server with 10+ tools. Load tested to ~1000 RPS.

  • Yandex
    2023–2025

    Senior Analyst-Developer

    Integrated LLMs into the automated ad moderation pipeline. Implemented LLM + RAG, reducing time-to-market by 2x. Trained and deployed 30+ classifiers to production. Enhanced fraud detection using new fingerprint features.

  • Gazpromneft Digital Solutions
    2022–2023

    Chief Analyst / Tech Lead NLP

    Developed a corporate text-to-SQL chatbot on RASA with a Python Action Server. Designed scalable architecture using NLP services and a knowledge graph. Implemented MLFlow/DVC for model tracking and deployed via Docker.

  • Gazpromneft Digital Solutions
    2019–2022

    Analyst → Lead Analyst

    Built a corporate knowledge graph (100K+ nodes) via text-mining 120 systems using NLP. Developed ETL processes on Airflow for data marts. Created analytics dashboards with textual query analysis.

  • Tinkoff Bank
    2016–2017

    Analyst

    Developed flexible interest rate calculation algorithms improving product NPV. Analyzed credit agreement rejection causes using text mining and NLP. Monitored key business metrics via Splunk.

Skills

  • Backend: Python 3.11+, FastAPI, asyncio, asyncpg, LangChain, LangGraph
  • Frontend: JavaScript (ES6+), React, HTML/CSS, Yandex Maps API, SSE
  • Data: PostgreSQL, pgvector, PostGIS, Redis, SQL
  • AI/ML: LLM (GPT-4, Claude, DeepSeek), RAG, embeddings, NLP, scikit-learn, TensorFlow
  • Infra & Tools: Linux, nginx, systemd, Docker, CI/CD, Cursor IDE, Git, MCP Protocol, LiteLLM

Projects

AiKarta
Visit site →

Real estate search platform with AI agent, built solo in 6 months using Cursor IDE + Claude.

  • Frontend:SPA on Vanilla JS, Yandex Maps API (~100K objects with clustering), AI chat with SSE streaming.
  • Agent API:Python/FastAPI (async), LLM agent (LangGraph/LangChain) with 3 execution strategies.
  • MCP Server:10+ tools, NLP query parsing, geodata (metro, districts, POI).
  • Knowledge Base:ETL pipeline, CIAN API parsing, embeddings (pgvector), price clustering.
  • Infra:Multi-model LLM Gateway (LiteLLM), monitoring dashboards, load tested to ~1000 RPS.

Contact