Available for AI Engineer / Automation roles · remote

OLEKSANDR POKARIEV

AI / Automation Engineer

Applied LLM Systems · RAG · Voice & Document AI · Workflow Automation · LLM Ops

Download CV (PDF) View Projects
✉ pokarev@gmail.com · ☎ Hover to reveal · LinkedIn · GitHub

About

Software engineer with 15+ years of commercial development, the last 2+ years focused full-time on applied AI / LLM systems. I design and operate production services around Large Language Models — RAG pipelines, real-time voice agents, document/OCR extraction, and multi-provider LLM integration with routing and fallback. I work by mixing ready-made tools with custom-built solutions: orchestrating pipelines in n8n / LangChain nodes where that ships fastest, and writing native code (Python / Node) where control and reliability matter. Full ownership: design → implementation → containerized deploy → monitoring & auto-heal. Primary language Python (FastAPI).

  • Base: Ostrówek, near Warsaw, Poland · remote (GMT+1/+2)
  • Languages: English — B1; Polish — B1; Ukrainian — Native; Russian — Native

AI / LLM Engineering

  • LLM pipelines & wrappers: model selection, provider-routing with automatic fallback, self-healing model watchdogs
  • RAG: embeddings, chunking, top-k retrieval, per-tenant namespaces — on Qdrant (n8n / LangChain nodes) and a from-scratch vector store (no framework)
  • Document / OCR AI: vision-LLM OCR + structured entity extraction with JSON-schema output and validation
  • Voice AI: Whisper STT → LLM → Silero / Piper / ElevenLabs TTS over WebSockets & Twilio
  • Prompt engineering: prompt-builder modules, structured-output prompting with retry; custom latency/token/quality bench
  • Self-hosting open models: vLLM, Ollama (Qwen / Llama / gemma)

Automation, Data & Ops

  • Workflow automation: n8n, LangFlow, LangGraph — triggers → actions, scheduled jobs
  • Data modeling: relational schemas + JSON/JSONB, idempotent UPSERT pipelines, multi-tenant structures
  • System integration: REST, webhooks, OAuth; Twilio, ElevenLabs, Stripe, Notion, marketplace & 1C/ERP
  • Backends: FastAPI, Node.js (Fastify/Express), REST, WebSockets, SSE streaming
  • DevOps: Docker, Linux, Nginx, PM2, WireGuard, Cloudflare Workers; AWS, GCP, Azure
  • Delivery: CI/CD via webhook auto-deploy, monitoring, watchdogs, auto-restart, documentation & handover

AI & Automation Projects

Flagship work from the last two years.

Live projects show a real screenshot. Click ↗ to open.

LiveTwinlix — AI voice-agent platform

Twinlix twinlix.com ↗

Multi-tenant AI voice-agent SaaS: inbound call handling, multi-channel (Twilio, Instagram, Facebook), knowledge management with document ingestion (6+ formats) synced to ElevenLabs, RAG on Qdrant, and Stripe billing.

Node / FastifyTypeScriptMySQL + Prisman8nQdrantElevenLabsStripe
LiveNEXX — B2B procurement platform

NEXX nexx.business ↗

Backend for a B2B procurement platform: relational data modeling for a multi-party marketplace (buyers / suppliers / catalog taxonomy), idempotent batch-UPSERT integration endpoints, and ERP sync (Syrve / 1C).

Python / DjangoPostgreSQLREST integrationsERP sync
Livegetqr.art — art QR generator

getqr.art getqr.art ↗

Art-QR generator with a vision-LLM step and a self-healing model watchdog that fails over when a provider model is retired. Live, free, no paywall.

NodeAstrovision-LLMmodel watchdog
LiveOrdora — marketplace store management + native-RAG CRM assistant

Ordora ordora.space ↗

Multi-tenant SaaS for marketplace store management, plus a CRM assistant with a hand-built vector-retrieval memory written from scratch — embeddings + cosine + a Float32 store and a search_memory tool doing top-k retrieval into an agentic tool-calling loop. No RAG framework.

Node.jsSQLDockercustom vector store
multi-persona LLM

One-One one-one.online ↗

Multi-provider LLM engine: per-slot model routing with automatic fallback, a dedicated prompt-builder module, persistent memory, and a voice fast-lane instrumented for latency/token benchmarking with per-call model override for A/B comparison.

Python / FastAPIvLLMOllamaDeepSeek
vision-LLM OCR

Document OCR & extraction service

Vision-LLM document service: /ocr for verbatim extraction and /receipt for structured JSON from receipt/invoice photos — key entities and attributes, with field validation and multi-sample verification.

Python / FastAPIGemini Visionstructured JSON
real-time voice

Callyou / Hermes voice loop

Browser and telephony voice agents: a real-time loop of Whisper STT → LLM → Silero / Piper / ElevenLabs TTS streamed over WebSockets, with barge-in handling.

Python / FastAPIWhisperSileroPiperElevenLabs
Liveiswish — embeddable Issues & Wishes board

iswish iswish.acebox.eu ↗

Hosted, multi-tenant Issues & Wishes board that embeds on any site with a single line. Multi-tenant data model on Supabase, deployed on Cloudflare.

NodeSupabaseCloudflare

// also built & running

KastaGrabData-sync & analytics: marketplace API + 1C → PostgreSQL (JSONB, idempotent UPSERT, scheduler) + owner dashboard.
RatioTelegram workflow bot: message/voice → parsed → scheduled jobs, with Notion integration. n8n + Python.
LanehubSelf-hosted Telegram hub-bridge routing conversations between people and AI agents. Python, Docker.
Puzzlore ↗Custom-puzzle generator with a frame/layer pipeline. Live domain. Node.

Technology

Primary language Python (FastAPI); also Node/TS, PHP, C#.
  • LanguagesPython, JavaScript / TypeScript, C#, PHP
  • Backend / APIsFastAPI, Node.js (Fastify / Express), REST, WebSockets, SSE streaming
  • LLM providersOpenAI, Anthropic (Claude API), Google Gemini (vision), DeepSeek, Mistral; self-hosted Qwen / Llama / gemma
  • RAG / vectorQdrant (n8n-orchestrated), custom embeddings + cosine vector store; LangChain via n8n nodes
  • Voice / visionWhisper, faster-whisper, Silero, Piper, ElevenLabs; Gemini Vision (OCR)
  • Serving / OpsvLLM, Ollama; health-monitoring, watchdogs, auto-heal, PM2
  • Automationn8n, LangFlow, LangGraph, custom workflows, cron / scheduled jobs
  • DatabasesPostgreSQL (incl. JSONB), MySQL / MariaDB, Prisma, Supabase
  • DevOps / CloudDocker, Linux, Nginx, WireGuard, Cloudflare Workers; AWS, GCP, Azure
  • IntegrationsTwilio, ElevenLabs, Stripe, Notion, Telegram, Google APIs, marketplace & 1C / ERP

Experience

2024 — Present
Founder & Lead AI Developer — AceBox (privacy-first AI systems)
Design and ship production LLM services across 5+ web projects: RAG on Qdrant plus a native vector-retrieval store; real-time voice agents (Whisper + TTS + Twilio); a vision-LLM document/OCR service; multi-provider LLM integration with routing, fallback and a self-healing model watchdog; self-hosting open models with vLLM & Ollama. Own the full lifecycle — design → Docker/Linux deploy → monitoring & auto-heal.
2017 — 2024
Senior / Lead Developer (C#, Unity) — production products
10+ years of large-scale production software — performance optimization, reliability and rapid prototyping; shipped 100+ mobile projects. The engineering foundation for the current AI-systems work.

Certifications

Anthropic Education — independently verifiable.

Earlier work — Game development

A decade of shipping, before AI.

10+ years as a Senior / Lead Unity (C#) developer. Hover a card to play a preview.

Crazy Robots Arena (2.5D)
SteamBot (3D quest)
Antari Studio (gameplay reel)
FreeKick Soccer
100+mobile projects shipped (Google Play / App Store)
10+ yrsUnity / C# production
DevGAMMKyiv — Most Innovative Game