Wiktor Jeleń

Backend & Platform Engineer

I'm Wiktor Jeleń, a backend and platform engineer based in Cracow, Poland. I build multi-tenant production systems, infrastructure and security-critical backends, and I wrote PythonMetaTrader5, an open-source library with over 44,000 PyPI downloads.

I design and build backend systems, infrastructure and production-grade platforms.

Cracow, Poland · root@yelon.dev · +48 576 706 766 · GitHub · LinkedIn · Download CV (PDF)

About Wiktor Jeleń →

Focus areas

Backend Systems

  • TypeScript
  • NestJS
  • Python
  • FastAPI
  • PostgreSQL
  • REST / OpenAPI

Platform & Infrastructure

  • Linux
  • Docker
  • Nginx
  • GitHub Actions
  • AWS
  • CI/CD

System Architecture

  • Multi-tenant isolation
  • Workflow orchestration
  • Failure modes
  • Integration & e2e testing

AI Integration

  • Multi-provider LLM layer
  • Model Context Protocol
  • RAG / vector search
  • LangChain / LangGraph
  • Evaluation harnesses
  • Media & speech pipelines

Projects

KONIK

Multi-tenant automation platform · Lead backend developer · 03–08.2026 · Architecture & delivery

Problem
An automation platform needs to coordinate external services and AI workflows while keeping each tenant’s data isolated.

Solution
Designed the TypeScript/NestJS backend with Temporal workflow orchestration and PostgreSQL row-level security. Built isolation-leak tests in CI, external-service integrations and n8n workflows, plus a multi-provider AI layer with an evaluation harness.

Stack

  • TypeScript
  • NestJS
  • Temporal.io
  • n8n
  • OpenAI / Claude APIs
  • PostgreSQL RLS
  • GitHub Actions

PythonMetaTrader5

Open-source order-execution library · 44,000+ PyPI downloads

Problem
The native Python API for MetaTrader 5 is low-level: raw retcodes, inconsistent data shapes, and no safeguards standing between a strategy and a live account.

Solution
A Python layer over MetaTrader 5 built for algorithmic trading: order normalization, full retcode handling and safeguards that keep execution stable against real capital.

Stack

  • Python
  • MetaTrader 5 API
  • PyPI

GitHub · PyPI

Supplier Integration Platform

B2B e-commerce client (under NDA) · 2026 · 59 → 109 suppliers · 830 tests · in production

Problem
A B2B e-commerce business buys from over a hundred suppliers, and each one exposes prices and stock differently: B2B panels on WooCommerce, PrestaShop, Shopify, Odoo and other platforms, file feeds, or nothing but an email inbox. And the ERP imports the result every hour, so one broken scraper can quietly shrink the catalogue.

Solution
Split the integration service out of a legacy system and grew it from 59 to 109 connected suppliers (89 domestic, 20 EU) covering 73,000+ products — I wrote 197 of its 207 commits. The platform runs two branches. Automatic suppliers: Playwright scrapers and API clients pull purchase prices and stock straight from their B2B panels. Manual suppliers, who have no panel: the system emails them through the Gmail API asking for current prices and stock, on a schedule built to protect the relationship — a person answers on the other side, so requests are paced instead of fired on every sync. Their replies come in whatever form that person chooses — figures typed in the email body, an XLSX or PDF attachment, or a photo of a document read with OCR — and an LLM turns each one into the same structure, which the customer-service team approves in an admin panel before it reaches the catalogue. Catalogues are published as SHA-256-checked versions with an atomic swap, and a drop of more than 20% blocks publication. CI audits secrets and system boundaries; deploys switch atomically and roll back automatically; 20 ADRs record the decisions.

Stack

  • Node.js 24
  • Express 5
  • MongoDB
  • Playwright
  • OpenAI API
  • Gmail API
  • GitHub Actions

RFQ Automation Service

B2B e-commerce client (under NDA) · 2026 · Built in 8 days · 445 tests · in production

Problem
The sales team sourced quotes by hand: read the customer request, guess which suppliers carry each item, write to each of them, then compare offers arriving in different units and currencies.

Solution
Designed and built from scratch in 8 days. The service takes a request from a salesperson by email, extracts the line items with an LLM into a strict JSON schema, and picks up to 25 suppliers from four data sources with a stated reason for each choice — 11 of 12 correct picks when replayed against historical quotes. It drafts the requests for approval and then handles the supplier replies by email, which arrive in whatever shape each supplier prefers: stock and price written in the body of the email, an XLSX attachment, or a photo of a document, which goes through OCR. Every reply is normalized to the same structure, offers are compared in net price per unit, and quantities are split between suppliers. Clean Architecture, DDD, ports & adapters and the Result pattern, with layer boundaries enforced in CI by dependency-cruiser. Security follows OWASP ASVS L2 and WSTG: 23 of 23 scenarios passed, Semgrep and secret scanning block merges, the recipient allowlist is enforced in code and senders are verified with SPF/DKIM.

Stack

  • Node.js 24 (ESM)
  • Clean Architecture / DDD
  • TDD
  • LLM structured outputs
  • Semgrep
  • dependency-cruiser

Marketing Content Generator

B2B e-commerce client (under NDA) · 2026 · Built in 5 days · ~390 tests · in production

Problem
Marketing had to publish a steady stream of social posts for a large, constantly changing catalogue, across several language markets, without promoting products that are out of stock or making claims EU consumer law forbids.

Solution
Built from scratch in 5 days and running in production, with a full Shoper integration: the tool reads the live catalogue — stock, bestsellers, promotions, product rotation — and plans a month of Facebook and Instagram posts from it. Graphics and videos cost nothing to produce: they are composed from templates, with product photos cut out by a background-removal model running locally on ONNX (BiRefNet), in four image formats plus a 10-second video per post. When a product is discounted, the layout adds a sale badge, the percentage off and the lowest price from the previous 30 days, as the Omnibus Directive requires. The OpenAI API is used only for text: Polish copy, then translation into 15 languages with product and market context, because a plain machine translator kept getting technical terms wrong; prices follow each region’s currency. Copy is also checked against Directive 2024/825 on green claims, and a batch generator handles up to 200 products per run. Jobs run through a pg-boss queue; CI/CD with automatic rollback and Google sign-in restricted to the company domain; 132 commits and 60 PRs.

Stack

  • TypeScript
  • Fastify 5
  • React 19
  • PostgreSQL 17 / Drizzle
  • pg-boss
  • Shoper API
  • OpenAI API
  • ffmpeg
  • ONNX Runtime

PIM & Dropshipping Takeover

B2B e-commerce client (under NDA) · 2026 · 0 → 12 suppliers · ~1.34M offers · in production

Problem
A product-information and dropshipping system inherited from a previous team, with no delivery pipeline, secrets committed to the code and margins that did not add up.

Solution
Reviewed the code, wrote down the technical debt, then shipped 144 commits and closed about 80 tasks in 3 days. Closed a critical hole — a public endpoint that issued an administrator token without logging in — rotated the JWT keys, removed hard-coded secrets and stopped an API key leaking through error messages. Fixed margin calculation (it was computed from the gross price), added protection against ordering twice from a supplier, live exchange rates from the national bank and daily order limits. Built 12 supplier integrations from zero (~1.34M offers), with product cards created automatically from Open Icecat. Built continuous delivery from zero: GitHub Actions, a least-privilege deploy account, automatic rollback, a pre-push test gate and worker health checks.

Stack

  • PHP
  • Laravel 12
  • Angular 21
  • MySQL
  • Redis
  • Docker
  • GitHub Actions

TradeLockerBot

Trading automation system · Docker image · CI/CD

Problem
TradingView can only fire a plain, unauthenticated HTTP webhook. Anything listening on a guessable URL is eventually found by a scanner — and this endpoint opens positions on a live account.

Solution
FastAPI service translating TradingView alerts into TradeLocker orders, with the webhook path generated as a random secret that is printed once at startup and never again. Shipped as a Docker image built by GitHub Actions.

Stack

  • Python
  • FastAPI
  • uvicorn
  • Docker
  • GitHub Actions

GitHub

Anatomy Project

Medical University of Silesia platform · 450+ concurrent users

Problem
An anatomy teaching platform had to serve a whole faculty at once, authenticate against the university directory, and stay up through exam-day traffic spikes.

Solution
NestJS 11 backend for 450+ concurrent users: LDAP/Active Directory authentication, JWT authorization, and an OpenAPI-documented REST API with pagination and filtering pushed down to the database.

Stack

  • NestJS
  • TypeORM
  • MySQL
  • LDAP
  • JWT
  • Jest

GEEETECH A10M Enhancement

Hardware R&D, Shenzhen · Shenzhen R&D

Problem
The A10M printer had structural and motion-control weaknesses that firmware tuning alone could not compensate for.

Solution
Structural redesign of printer components in Fusion 360, combined with Klipper firmware optimization and improved stepper motor control.

Stack

  • Fusion 360
  • Klipper
  • C++
  • CAD

IoT Pipe Monitoring

ThingsBoard integration for the heating industry · AWS alerting

Problem
District-heating operators find out about pipe degradation after a failure. The impedance data existed, but never reached anyone who could act on it.

Solution
Impedance monitoring built on ThingsBoard: embedded C++ firmware feeding a web interface, with AWS-based alerting on threshold breaches.

Stack

  • ThingsBoard
  • C++
  • AWS
  • Embedded

Experience

Backend Engineer (contract)

B2B e-commerce automation (under NDA) · 09.2026 - 11.2026 · Poland, Remote

  • Extended a supplier-catalogue integration platform from 59 to 109 supplier integrations covering 73,000+ products: automatic suppliers scraped from their B2B panels (WooCommerce, PrestaShop, Shopify, Odoo and others) with Playwright or read through APIs, and manual suppliers emailed for current prices and stock on a schedule that protects the relationship, whose replies (email text, XLSX, PDF or a photo read with OCR) are parsed by an LLM and approved by the team in an admin panel. Catalogues are published as SHA-256-checked versions with atomic swaps, and a drop of more than 20% blocks publication; 830 automated tests.
  • Designed and built from scratch in 8 days an RFQ automation service: it takes a request from the sales team by email, extracts line items with an LLM into a strict JSON schema, matches up to 25 suppliers from four data sources and prepares the requests; supplier replies with stock and price — written in the email body, sent as XLSX or as a photo read with OCR — are normalized, compared per unit and quantities split between suppliers. Clean Architecture, DDD, ports & adapters and TDD with layers enforced in CI; 445 tests including golden tests on real requests and 13 prompt-injection scenarios against the live model; security to OWASP ASVS L2 / WSTG with 23 of 23 scenarios passed, Semgrep and secret scanning gating every merge.
  • Built from scratch in 5 days a marketing-content generator now in production (TypeScript, Fastify, React, PostgreSQL, pg-boss) with a full Shoper integration: it plans a month of posts from the live catalogue and composes images in four formats and a 10-second video from templates at no generation cost, adding a sale badge, the discount and the 30-day lowest price (Omnibus) for discounted products; the OpenAI API writes the Polish copy and translates it into 15 languages with product context; ~390 tests, CI/CD with automatic rollback.
  • Took over a legacy PIM / dropshipping system (PHP / Laravel, Angular, MySQL, Redis, Docker): reviewed the code and documented the technical debt, closed a critical vulnerability where a public endpoint issued an admin token without login, rotated JWT keys, removed hard-coded secrets, stopped an API key leaking through error messages, fixed margin calculation, added protection against duplicate supplier orders, and built 12 supplier integrations from zero (~1.34M offers). Set up continuous delivery from scratch with a least-privilege deploy account and automatic rollback.
  • Integrated the OpenAI and Claude APIs for the LLM features, and the company Google Workspace through Google OAuth for the email-driven workflows.
  • All four systems run in production, with 1,600+ automated tests, CI/CD with automatic rollback, and secret scanning and Semgrep gates on every merge.

Lead Backend Developer

KonikSystems · KONIK · 03.2026 - 08.2026 · Remote

  • Led the architecture and development of a multi-tenant automation platform (TypeScript / NestJS, PostgreSQL, Temporal.io, React dashboard) and a team of two developers, while remaining the primary contributor to the codebase.
  • Designed tenant isolation in PostgreSQL using row-level security, with automated isolation-leak tests running in CI.
  • Built the test and delivery pipeline: unit and integration tests, Temporal TestWorkflowEnvironment, Playwright, and a multi-stage GitHub Actions CI/CD.
  • Designed the integration layer: Stripe, Google Ads/Analytics, Meta API, social publishing, scraping and enrichment, n8n workflows, and a video processing pipeline.
  • Built the LLM and generative-AI integration layer — Claude, OpenAI (GPT) and Gemini APIs, MCP, Flux/LoRA and Whisper — together with the evaluation harness for AI output quality.

Backend Engineer

Randlab Software · Anatomy Project · 09.2025 - 02.2026 · Remote

  • Designed and deployed a scalable backend (NestJS 11, TypeORM, MySQL 8) for 450+ concurrent users of a Medical University of Silesia platform.
  • Integrated authentication with the university Active Directory (LDAP) and JWT-based authorization; hardened the API with OWASP Top 10 audits, strict input validation (Zod) and data sanitization.
  • Designed a REST API with a Swagger/OpenAPI contract — pagination, filtering and sorting pushed down to the database rather than into application memory — plus unit and integration tests in Jest.
  • Set up production environments (Ubuntu, Nginx reverse proxy, SSL, PM2) and CI/CD pipelines in GitHub Actions.

R&D Systems Engineer

GEEETECH · 08.2024 - 08.2025 · Remote · Shenzhen R&D team

  • Owned system integration and Klipper firmware optimization in Python for the Shenzhen R&D hub, working fully remotely and self-directed.
  • Redesigned structural components of the A10M printer in Fusion 360 and optimized stepper motor control.
  • Coordinated technical requirements between European stakeholders and the Shenzhen R&D team.

Sales Systems & Automation Specialist

Randlab Software (Polsat Plus Group) · 04.2023 - 07.2024 · Remote

  • Automated client documentation generation and configured complex sales offers for POS, telemarketing and mobile sales channels.

Backend & IoT Developer

Randlab Software · 09.2022 - 03.2023 · Nowy Sącz, Poland

  • Built the Node.js backend for a sensor-based pipe impedance monitoring system for the heating industry on the ThingsBoard IoT platform.
  • Integrated hardware (C++) with web interfaces and implemented AWS-based alerting.

Technical skills

Backend & Data

  • TypeScript
  • Node.js
  • NestJS
  • Temporal.io
  • PostgreSQL
  • Redis
  • MySQL
  • Python
  • FastAPI
  • Django / DRF
  • PHP / Laravel
  • SQLAlchemy
  • MongoDB

Platform & Infrastructure

  • Git (GitHub, GitLab, GitHub Actions)
  • Linux
  • Docker
  • Kubernetes
  • Terraform
  • Nginx
  • AWS
  • GCP
  • Azure
  • Hetzner
  • CI/CD
  • n8n
  • Google Workspace

Security

  • PostgreSQL RLS
  • IAM
  • LDAP / AD
  • JWT
  • OWASP Top 10
  • TLS

AI & Automation

  • Claude / OpenAI / Gemini APIs
  • Model Context Protocol
  • RAG / vector search
  • LangChain / LangGraph
  • Evaluation harnesses
  • Flux / LoRA
  • Whisper

Additional

  • React
  • Next.js
  • Vue.js
  • C++
  • Embedded systems
  • Fusion 360

Consulting

Paid Client Work — Embedded Linux, Firmware & Hardware

Alongside my software and infrastructure work, I take on independent technical consulting projects involving embedded Linux, Armbian, Klipper, Marlin and 3D-printer hardware. 32 public reviews since 2022, 31 of them at five stars, from clients in ten countries. Every review is publicly verifiable on Fiverr.

  • 32 public reviews
  • 31 of 32 five-star
  • 10 countries

This guy knows what he is doing. Helped me get my modified ender 3 printing again. I have been scratching my head on the problems with it for months and he helped me get it working in a couple days.

ccrank85 · US

Was very helpful and willing to help immediately. Was also very flexible in helping with the issue we were in vastly different time zones. 10/10 would reach out for consulting again.

skilnotfound · US

Verify on Fiverr

Interests & R&D

I explore evolutionary computation for algorithmic trading: generating strategy populations in StrategyQuant X, then stress-testing their robustness before live execution.

  • Machine Learning
  • Genetic Algorithms
  • Evolutionary Computation
  • StrategyQuant X
  • Algorithmic Trading

Education

B.Eng. in Computer Science, Cybersecurity — in progress (part-time)

WSB-National Louis University · Nowy Sącz, Poland · 10.2024 – Present

Mechatronics Technician

Technical School of Electrical Engineering and Mechanics · Nowy Sącz, Poland · 2019 – 2024

Languages

  • Polish — native
  • English — working proficiency (EU–China R&D collaboration, international clients)

Elsewhere

Contact

Open to backend, platform and system-architecture work — remote or Cracow, Poland.

root@yelon.dev · +48 576 706 766

More about Wiktor Jeleń