AI AUTOMATION ENGINEER

Gleb Pyatkov

Agentic Workflows & Multi-Agent Systems

I architect and ship production AI systems — multi-agent orchestration, RAG and vector search, fine-tuning and prompt engineering — together with the surrounding software: web interfaces, native macOS (Swift) applications, backend services and infrastructure. AI-native, full-cycle from architecture to operations, delivered solo.

Remote-first · Open to relocation

email telegram download cv

Python
LangChain
LangGraph
RAG
pgvector
Multi-Agent Systems
FastAPI
Prompt Engineering
Claude Agent SDK
MCP

3+ yrs DevOps & infra

100% solo, full-cycle

Web + macOS full-stack + native

95% automated

24/7 autonomous

6 in production

// what I build

Projects

Seven production systems — AI agents, automation pipelines and full-cycle SaaS, all shipped solo.

Whispo

whispo.app ↗

Cross-platform voice-dictation SaaS — native macOS app, FastAPI backend, web portal

Native macOS app (Swift/SwiftUI, ~12.5k LOC): global push-to-talk hotkeys, Groq Whisper transcription, clipboard + ⌘V injection into any app, cloud-synced Memory Storage
FastAPI backend — 15 API modules: Supabase JWT (ES256/HS256), Redis per-user/IP rate limiting, prompt-injection-hardened Groq Llama-3.3-70B text restructuring
End-to-end monetization — two payment gateways (YooKassa + CryptoCloud) with auto-renewal and referral bonuses; Prometheus/Grafana/Loki observability, Docker + nginx + Cloudflare

macOS + Web JWT ES256 Payments

stack Swift · SwiftUI · React · TypeScript · FastAPI · Supabase · Redis · Groq (Whisper + Llama 3.3) · YooKassa · CryptoCloud · Docker

Hookora

FLAGSHIP

TS monorepo for short-video creators: IG→DM funnels, viral search, AI captions, autopost

4 independent Fastify 5 + TypeScript services over one PostgreSQL 16 behind an nginx gateway (Docker Compose); each self-migrates and returns a uniform {success, data, error} envelope
Production AI: Groq Whisper-large-v3 transcription + Llama-3.3-70B caption generation, with system/user prompt-injection isolation over untrusted transcripts
End-to-end social integrations — OAuth publishing to YouTube, TikTok & Instagram (tokens AES-256-GCM at rest), Meta-webhook comment→DM funnels (HMAC-verified), viral search via YouTube Data API + Apify

4 services 782 tests AES-256-GCM

stack TypeScript · Fastify 5 · React 18 · Vite · Tailwind v4 · PostgreSQL 16 · Groq (Whisper + Llama 3.3) · Remotion · Docker · Nginx

HH Job Agent

CURRENT

Autonomous AI agent that chats with recruiters and applies to jobs for me

Claude-based agent across 5 conversation channels (AI recruiter, live recruiter, Telegram, Google Forms, hh.ru forms)
RAG retriever over a modular knowledge base — only relevant cards enter the prompt; auto-learns from clarifications
Autopilot scheduler, Playwright over SOCKS5, a Telegram userbot (GramJS) and an admin control bot

5 channels RAG Autopilot

stack Node.js · TypeScript · Playwright · GramJS · Claude (MCP, RAG) · Telegram Bot API

MBand

Internal AI platform automating image, video & copy production for a fashion brand

Solo full-cycle: FastAPI backend — 23 route groups, ~200 endpoints, 30+ service modules; ~68K LOC across Python + TypeScript
Orchestrates 6+ AI providers (Gemini, OpenRouter, Kling, Veo, Topaz, Seedance) behind a persistent job queue with parallel processing, per-type concurrency limits and stuck-job recovery on restart
React + TypeScript SPA — 19 tool pages (Canvas crop editor, drag-&-drop, SSE-streamed AI chat) on Supabase (Postgres + Auth + Storage), with Prometheus/Loki/Grafana observability

~200 endpoints 6+ AI providers −75% time

stack Python · FastAPI · React · TypeScript · Vite · Supabase · FFmpeg · Gemini · OpenRouter · Kling · Topaz · Docker

VikiShop

Hands-off PUBG UC resale — bots buy on Midasbuy, auto-sell across 3 marketplaces

Microservice Docker stack on one PostgreSQL — 3 marketplace bots (FunPay/Playerok/GGSEL) + Midasbuy purchase worker + Telegram admin bot + FastAPI/React CRM + Prometheus/Grafana; services talk only via a DB queue
11-state order FSM with a validated transition graph and retry/escalation; concurrency-safe via SELECT … FOR UPDATE SKIP LOCKED queue claims and row-locked transitions; idempotent on a unique platform order id
Stealth browser automation on patchright (anti-detect Playwright fork) bypassing Midasbuy's anti-bot — per-account browser pool, multi-format IMAP OTP (3DS/EPN), priority card failover, Fernet-encrypted card data

11-state FSM 3 marketplaces ~495 tests

stack Python · asyncio · SQLAlchemy 2.0 · asyncpg · PostgreSQL · patchright · FastAPI · React · Docker · Prometheus

Bybit Trading Bot

Event-driven Bybit USDT-perp futures bot — short-only fade-pump on Telegram signals

Dual Bybit v5 WebSocket: a private order/execution/position stream with manual HMAC-SHA256 auth as the source of truth, plus a public ticker feed — orjson serialization, auto-reconnect, graceful SIGINT/SIGTERM shutdown
Exchange-side risk management: server stop-loss attached atomically to the entry order (no unprotected window), leverage control with TTL cache + retries and Bybit error-code handling, fixed-notional sizing rounded to the lot step
Persisted trade ledger on PostgreSQL + SQLAlchemy 2.0 (typed Mapped) + Alembic: a trade_log table with a 5-state trade FSM, partial-fill aggregation, weighted entry price and PnL via ORM hybrid-properties

dual WebSocket HMAC-SHA256 FSM ledger

stack Python 3.11 · asyncio · uvloop · pybit · websockets · aiogram · SQLAlchemy 2.0 · Alembic · PostgreSQL · Docker

HyperEVM Liquidity Bot

Concentrated-liquidity market-making bot on HyperEVM (DeFi)

Auto-recenters an ultra-narrow LP window every cycle (default 5s) — closes (decreaseLiquidity → collect → burn) and re-mints on the live tick when price exits range or nears an edge, keeping the position earning fees 24/7
Low-level on-chain Ethereum: raw eth_call to slot0 with manual two's-complement int24 tick decode, keccak-selector revert decoding and tickSpacing-aligned range math — abstracting two CL DEX forks (Project X, Ramses)
Telegram-orchestrated 24/7 service: synchronous web3 loops in daemon threads wrapped as asyncio tasks, a Redis-persisted task registry that auto-restores after restart, plus a 3-asset inventory balancer routing swaps via KyberSwap/OpenOcean

auto re-centering Uniswap V3 forks on-chain CL math

stack Python · web3.py · python-telegram-bot · Redis · PostgreSQL · KyberSwap · OpenOcean · HyperEVM

// how I work

Approach

An engineer who takes business processes from idea to production with AI agents. I orchestrate LLM agents (Claude Agent SDK, MCP, RAG) and own the whole stack — backend, integrations, frontend, infrastructure and deployment.

AUTOMATION-FIRST

I design processes around AI agents and multi-agent orchestration, automating up to 95% of manual work.
SOLO, FULL-CYCLE

From architecture and design to deploy and monitoring — including payments and auth.
AI-NATIVE DEVELOPMENT

LLM agents at every stage: design, coding, testing, code review.
PRODUCTION-READY

Every project ships: Docker, health checks, monitoring, error handling, scale.

// stack

Skills & tooling

The stack I use to design, build and ship autonomous systems — end to end.

ai & agents: AI AutomationMulti-Agent SystemsAgentic WorkflowsLLM AgentsLangChainLangGraphLlamaIndexRAGVector SearchEmbeddingspgvectorFine-tuningLoRA / QLoRAHugging FacePyTorchTransformersLLM EvaluationAgent HarnessClaude Agent SDKModel Context Protocol (MCP)Prompt Engineering
backend: PythonFastAPIasyncioSQLAlchemyPydanticNode.jsTypeScriptFastifyREST APIsGraphQLgRPCWebSocketsServer-Sent EventsMicroservices
data & infra: PostgreSQLpgvectorPinecone / QdrantRedisDockerdocker-composeKubernetesNginxCaddyCI/CDAWSPrometheusLinux / VPS
frontend & software: ReactNext.jsTypeScriptViteTailwind CSSZustandReact QuerySwiftSwiftUImacOS apps
automation: PlaywrightTelegram Bot APIGramJSIMAPGoogle Sheets APIFFmpeg

// experience

Experience

Mario Berlucci

Fullstack Developer

Jan 2024 — Present

Footwear & apparel retail brand

Built a content-automation platform that cut video processing time by 75% (20+ min → 3–5 min) and automated 95% of manual operations
Designed a multi-service architecture — 20 backend services with real-time WebSocket/SSE progress
Integrated multiple external AI services with load-balancing, retry logic and dynamic pricing
Built a fault-tolerant job queue (memory → DB → file) with auto-recovery and parallel processing of up to 4 videos
Set up production deployment — Docker, Caddy reverse proxy with SSL, rsync delivery, health checks

$ contact

Let's build something that runs itself.

Open to AI Automation / Agentic Engineering roles — remote-first, open to relocation.

email [email protected] telegram @pyatkovgleb phone +7 929 757-39-01 english B2 · Password Skills Plus →

Gleb Pyatkov

AUTOMATION-FIRST

SOLO, FULL-CYCLE

AI-NATIVE DEVELOPMENT

PRODUCTION-READY

Mario Berlucci