Sergey Makovsky

Designed an AI system of agents to automate business processes (200+ users across hotel booking and CRM workflows via Bitrix24, RetailCRM, Yclients). Built an in-house framework for automating integrations with external APIs, reducing time-to-production for a new integration from 2–3 weeks to 2–3 days. Designed RAG systems on vector databases (Qdrant, Faiss) and optimized work with LLMs (OpenAI, Qwen, YandexGPT, DeepSeek) to reduce token consumption, enable response caching, and implement prompt compression. Developed tooling for prompt engineering and orchestration of LLM calls (function calling, structured outputs, agent chains).

Backend for a mobile app for cycling routes: migrated routing to a self-hosted tool (Valhalla) reducing per-user costs by 25–30%. Optimized REST API performance with Redis caching and asynchronous Celery tasks. Built a system for storing and analyzing geodata (PostGIS, heatmap visualization). Maintained code quality with 85% test coverage, CI/CD pipelines, and Docker containerization.

Developed text and voice B2B bots for Telegram, WhatsApp, and SberSalut with custom NLU modules (50+ intents; up to 20 requests/sec). Deployed a voice assistant for a call center with Asterisk integration (up to 60% inbound calls handled without an operator). Created an AI financial-analytics service with automatic report generation and Q&A based on user data. Built MLOps infrastructure with Docker images and GitLab CI pipelines for automated deployment.