# Oliwier Data > Full-Stack Developer & AI Engineer. I build production web applications, AI agents grounded in company data, and the infrastructure they run on. One person takes a product from idea to a live deployment: the application, the AI agents and automation around it, and the infrastructure underneath. - Rzeszów, Poland · Remote - Open to full-stack & AI engineering freelance projects - 20+ n8n automation workflows running in production - 10+ AI agents & RAG systems shipped — OpenAI, Claude, Gemini - 30+ repositories: commercial, product and own projects, most under NDA - 100+ users supported on the production infrastructure I maintain ## Three engineering pillars ### Full-Stack Web Engineering Fast, modern web applications, internal systems and conversion platforms — built end to end, from the database to a polished frontend. Where it runs: - 30+ production and commercial repositories - Clients in logistics, hospitality and healthcare-adjacent sectors Tools: - Next.js 16 - React - TypeScript - Prisma ORM - PostgreSQL - SQL - Tailwind CSS - Three.js / WebGL - Playwright - REST APIs - Node.js ### AI Engineering: RAG Systems, Voice Agents & Process Automation AI agents grounded in a company's own data (RAG), reached by chat or by voice, and multi-step automation pipelines that replace manual work across a team. 20+ workflows in production: - Lead intake, qualification and routing - AP invoice processing with OCR and Sheets - Zendesk ticket triage and categorisation - Data sync, automated reporting and error handlers with alerting Tools: - OpenAI API - Claude API - Gemini API - RAG architecture: chatbots and assistants grounded in client documentation, with conversation memory and tool calling - Vapi & ElevenLabs: voice agents: a spoken question goes through the same retrieval, memory and tool calling - n8n: LangChain-powered agent nodes and custom workflows - Python - REST API integrations - OCR - Google Sheets API - Zendesk ### DevOps, Infrastructure & Self-Hosting Containerisation, virtualisation and automated deployments that keep the applications and AI agents I build stable and secure in production. In production: - Infrastructure for 100+ users, maintained day to day - Every app and agent above ships as a container through a pipeline - Self-hosted services published through Cloudflare Tunnel, with no inbound ports open - Encrypted backups, with restores rehearsed Tools: - Docker & Docker Compose: multi-container architecture in production - CI/CD pipelines: automated builds, end-to-end test gates and deployments - Cloudflare Tunnel & Nginx: services published through a tunnel and a reverse proxy, with no inbound ports open - Proxmox VE: server virtualisation, virtual machines and clusters - Linux & Windows Server: administration, service upkeep, TCP/IP, DHCP, DNS, diagnostics ## Systems in production ### Artur Express Full-Stack Web Engineering. Platform and conversion funnel for a logistics carrier. Challenge: A US trucking carrier needed its website rebuilt as a lead engine: every visitor path should end in a captured, routable enquiry — and the pages had to load fast. Built: End-to-end UX/UI design and build in Next.js, TypeScript and Tailwind CSS: a multi-path conversion funnel with lead capture, pay calculators, a news system and a JWT-secured admin CMS on PostgreSQL + Prisma — shipped as Docker images through a CI/CD pipeline that runs the Playwright suite before every deploy, and published through a Cloudflare Tunnel with no inbound ports open. Results: - Lighthouse 98 / 100 / 100 / 100 - LCP 0.8 s · CLS 0.004 - 62 pages · 42 content types - 19 Playwright end-to-end tests, run before every deploy Stack: Next.js 16, React 19, TypeScript, Tailwind CSS, PostgreSQL, Prisma, Playwright, Docker, CI/CD, Cloudflare Tunnel - [On the page](https://oliwierdata.pages.dev/#case-artur-express) - [Visit the live site](https://arturexpress.com) ### Enterprise RAG & AI Agents AI Engineering. Assistants that answer from a company's own documentation. 10+ production deployments. Challenge: Teams lose hours answering questions that are already written down — in documentation nobody has time to search, and that a generic chatbot knows nothing about. Built: AI assistants and chatbots powered by clients' internal documentation, built on the OpenAI, Claude and Gemini APIs with n8n's LangChain agent nodes: conversation memory, retrieval over RAG knowledge bases and integration with external tools. The same agent also takes a voice channel, built on Vapi and ElevenLabs: a spoken question runs through the same retrieval, memory and tools, and the answer is spoken back. Results: - Answers grounded in the client's own documents, not model guesswork - Conversation memory and tool calling in every agent - One agent behind chat and voice: Vapi carries the call, ElevenLabs the speech - Self-hostable end to end: n8n, pgvector and Postgres in Docker Stack: OpenAI API, Claude API, Gemini API, n8n, LangChain agent nodes, Vapi, ElevenLabs, pgvector, PostgreSQL, Docker Workflow: Google Drive + Chat + Voice call -> Embeddings -> pgvector store -> RAG agent -> Postgres memory + Answer Connections: Google Drive -> Embeddings; Chat -> Embeddings; Voice call -> Embeddings; Embeddings -> pgvector store; pgvector store -> RAG agent; RAG agent -> Postgres memory; RAG agent -> Answer - [On the page](https://oliwierdata.pages.dev/#case-rag-ai-agents) - [Reference implementation on GitHub](https://github.com/oliwier-xiao/n8n-rag-ai-agent) ### Production Automation Pipelines AI Engineering & Automation. Business processes that run themselves, with a human where it matters. 20+ workflows in production. Challenge: Invoices, support tickets and sales leads were handled by hand: copied between systems, answered late and dropped without anyone noticing. Built: Automated business processes in n8n, Python and REST APIs: OCR-based AP invoice processing, Zendesk ticket triage, sales-lead routing, data sync between systems and central error monitoring with alerts. Results: - Leads: signature-verified, de-duplicated, scored 0–100 and routed - Invoices: OCR, AI extraction, approval in Slack, register in Sheets - Tickets: AI drafts the reply, a human approves before it is sent - Every failure lands in an ops channel — nothing fails silently Stack: n8n, Python, REST APIs, OCR, Google Sheets API, Zendesk, Slack Workflow: Gmail intake -> OCR -> LLM extraction -> Dedupe + PO + vendor match -> Slack approval -> Sheets register + Error alerts Connections: Gmail intake -> OCR; OCR -> LLM extraction; LLM extraction -> Dedupe; LLM extraction -> PO + vendor match; Dedupe -> Slack approval; PO + vendor match -> Slack approval; Slack approval -> Sheets register; Slack approval -> Error alerts - [On the page](https://oliwierdata.pages.dev/#case-automation-pipelines) - [Hire me for automation](https://www.fiverr.com/oliwier_data/build-n8n-workflow-automation-and-ai-agents) ### Containerized Infrastructure & Commercial Web Apps Full-Stack & DevOps. The platform everything above is deployed on. 30+ repositories. Challenge: An application or an AI agent is only worth something while it is up: deployed the same way every time, backed up, and recoverable when a server fails. Built: Web applications (Next.js 16, Prisma, PostgreSQL) and internal automation systems deployed as Docker Compose stacks on Linux and Proxmox, shipped through CI/CD pipelines that run end-to-end tests before a deploy — with a reverse proxy and SSL in front, a Cloudflare Tunnel so that no inbound port is open, and encrypted, automated backups behind. Results: - Multi-stage Docker images and Compose stacks - Jenkins and GitHub Actions pipelines, with end-to-end tests gating the deploy - Cloudflare Tunnel in front: no inbound ports open on the host - Encrypted, automated backups with integrity checks and rehearsed restores - Infrastructure serving 100+ users Stack: Docker Compose, Proxmox VE, Linux, CI/CD, Nginx, Cloudflare Tunnel, Next.js 16, PostgreSQL Workflow: Git push -> CI + E2E tests -> Docker build -> Compose stack -> Nginx -> Cloudflare Tunnel + Backups + restore Connections: Git push -> CI + E2E tests; CI + E2E tests -> Docker build; Docker build -> Compose stack; Compose stack -> Nginx; Nginx -> Cloudflare Tunnel; Nginx -> Backups + restore - [On the page](https://oliwierdata.pages.dev/#case-containerized-infrastructure) - [GitHub profile](https://github.com/oliwier-xiao) Most of the 30+ repositories are under NDA: commercial work for logistics, hospitality and healthcare-adjacent clients. Ask, and I will walk you through what I can show. ## Follow one through ### An invoice arrives A supplier emails a scanned PDF invoice to accounts payable. 1. Gmail intake: The message is picked up and the attachment pulled out. 2. OCR: A scanned PDF is read into text; a digital one skips this step. 3. LLM extraction: Supplier, amounts, dates and line items become structured fields. 4. Duplicate & PO check: The invoice is compared with earlier ones and matched to its purchase order and vendor. 5. Approval in Slack: A person sees the extracted fields in an editable form and approves or corrects them. 6. Sheets register: The approved invoice is written to the register. Filed and approved by a person; any failure alerts the ops channel instead of passing silently. ### A lead comes in A visitor submits the enquiry form on a client's website. 1. Signed webhook: The site sends the lead with an HMAC-SHA256 signature. 2. Signature check: A request that does not verify is rejected before anything else runs. 3. File and de-duplicate: The lead is stored and matched by email, so the same person is never filed twice. 4. Score 0–100: The lead is scored on what it says and how complete it is. 5. Route: Hot goes to Slack and email, warm to Slack, cold is filed. Every lead is stored and the right person hears about the good ones at once. ### A support ticket opens A customer opens a new ticket in Zendesk. 1. New ticket: Zendesk hands the ticket to the workflow as it is created. 2. Classify: An AI agent sorts it by topic and urgency. 3. Draft a reply: The agent writes an answer grounded in the help-centre articles. 4. Human approval: A person approves or edits the draft in Slack before anything is sent. 5. Send or escalate: The reply goes to the customer, or the ticket is passed on for manual review. Customers get a grounded answer faster, and nothing reaches them without a person seeing it. ## Where I work ### Full-Stack Developer & AI Engineer, Freelance & commercial projects Sep 2025 — present · Remote End-to-end delivery of Next.js 16 applications (Prisma + PostgreSQL, Tailwind, Docker), 10+ AI and RAG agents on OpenAI, Claude and Gemini, and 20+ production automation workflows in n8n. ### IT & Infrastructure Specialist (L1/L2) · Internal Automation, AmeriPol Mar 2026 — present · Rzeszów Maintaining Docker Compose infrastructure on Linux, Proxmox virtualisation and CI/CD pipelines; building internal automations in n8n, Python and REST APIs that cut manual work; technical and network support for 100+ users. ## Have something to ship? Send a short brief — what you need built, the stack you are on and the timeline — and I will come back with how I would build it. - [Email](mailto:oliwier.data@gmail.com): oliwier.data@gmail.com - [LinkedIn](https://www.linkedin.com/in/oliwier-data): linkedin.com/in/oliwier-data - [Website](https://oliwierdata.dev): oliwierdata.dev - [GitHub](https://github.com/oliwier-xiao): github.com/oliwier-xiao - [Fiverr](https://www.fiverr.com/sellers/oliwier_data): fiverr.com/oliwier_data