Production AI systems — RAG pipelines, content generation, and internal AI tooling. Built end-to-end. Measured in production.
Based in Moscow · open to remote or relocation across EU/US timezones. Python backend since 2023 (ETL, data pipelines, internal services), moved into applied AI in January 2025. Now over a year as AI engineer at Hao — shipping production RAG and content generation pipelines. Graduating Bauman MSTU (CS) in 2026.
AI-autonomous orchestrators. Led the video generation pipeline end-to-end and shipped the marketing site. Scope: dialogue generation, TTS integration, image compose, cost instrumentation, RAG work.
Led the rebuild of a linear single-path orchestrator into a two-stage CLI with JSON contract between stages. Stage 1 runs an eight-step dialogue pipeline with an optional human-review seam, against per-character YAML profiles. Stage 2 runs TTS and pluggable image generation in parallel, composes vertical video via FFmpeg with sprite overlays and ASS karaoke subtitles, uploads to S3. Built-in per-generation cost instrumentation, typed config, structured logging, rate-limited retries on external APIs.
6× faster processing · −60% generation cost · batch mode enabled continuous output
Metrology company · internal tooling. Owned Python services covering ETL and data quality across internal catalogs.
Automated manual moderation across product catalogs and added validation across data imports from partner systems.
3× faster site moderation · −60% import errors
Daniil joined the team while still a student and within months was shipping ETL services we still run today. Rare to see that level of ownership and follow-through early on.
— Yury · Team Lead, Zenit-Elektro
A Chromium side-panel agent that runs tasks inside your real browser session — no headless cloud VM, no credential copy-paste. Built on a LangGraph.js Plan-and-Execute StateGraph with per-tool budgets, tab isolation, and a tool-family-aware duplicate guard that recovers from antibot click loops by switching to URL-based navigation or HITL hand-off.
AI assistant for personal and organizational networks. Multi-tenant — private and open communities connect their contacts with per-tenant RLS isolation. Semantic search over embeddings, news parsing via Jina AI, reranking loop grounds answers in source records.
Personalized Telegram news digest. Scrapes configured channels, generates an AI summary tailored to the user's focus and profile. Two-agent architecture: stateless deterministic pipeline + stateful conversational agent with persistent LangGraph checkpointing over Supabase.