Freelance AI engineer · Budapest · DE/EN/RU/HU · remote
Sergej Kukshausen AI skills, agents and MCP servers for engineering teams
I build custom AI skills, agents and MCP servers for Claude, GitHub Copilot and Codex, and use them every day in real delivery: an upgrade of a ten-year-old M/Text platform, run with GitHub Copilot, process automation across five teams with Claude Code, and the MCP server behind this website. 11+ years of software delivery, the last eight in banking and fintech.
Most recently I used GitHub Copilot to upgrade an M/Text platform that had grown over ten years, together with the three core systems that depend on it, including the Apache load balancer. See the work
Open to new engagements · Updated
How I can help
Four things I do for engineering and operations teams, remote from Budapest or on-site in the DACH region.
AI skills, agents & MCP servers
Custom skills and agents for Claude, GitHub Copilot and Codex, MCP servers that connect AI assistants to your systems, and process automation built with Claude Code. Integrated with Jira, Confluence, GitLab and n8n.
Shipped in 2026: an invoice-checking AI (6,755 jobs, 99% exact matches), a live bank-matching tool and two MCP servers, one for webshop prices and the one behind this site. Rollout, guardrails and training included.
Modernise banking document workflows and output management, and upgrade long-running M/Text platforms with AI assistance.
Done at Capgemini: 75% less manual processing, about €180K saved per year. In 2026: an upgrade of a ten-year-old M/Text platform, run with GitHub Copilot.
Automation consultant for an online printing group in three countries. I shadow the people doing the work, then use Claude Code as my daily build partner: meeting transcripts become numbered requirements, requirements become clickable prototypes the same day, and prototypes become specs the in-house developers code from. Delivered an AI tool that reads supplier invoices and checks them against 6,755 production jobs with 99% exact matches, a bank-matching tool now live on the shop system, and an MCP server that lets AI assistants quote live prices from the webshops.
Upgrading a ten-year-old M/Text platform with GitHub Copilot
Led the upgrade of a business-critical M/Text document platform that had grown over about ten years, together with the three core systems that depend on it, including the Apache load balancer. GitHub Copilot in VS Code was my assistant throughout: an ordered change plan with dependencies and approval points, operational scripts, log correlation while troubleshooting, AI-assisted checks of backups, deployment readiness and processing results, and the status updates and handovers. Every production decision and approval stayed with me. The upgrade went live with business workflows validated and rollback assets kept.
Led the modernization of a high-availability payment routing system from on-prem mainframe to cloud-native microservices on AWS: 31% lower infrastructure costs, 12% faster incident resolution, PSD2/GDPR-compliant throughout.
Python · AWS · Kafka · Kubernetes · GitLab CI/CD
FISERV · TEAM LEAD
A team that stays and ships
Lead a cross-functional team of engineers and analysts: hiring, 1:1s, coaching and performance reviews. Introduced Scrum across teams and cut product launch time by 21%, with three audits passed at zero findings.
Scrum · Product partnership · Hiring · Coaching
FISERV · HANDS-ON
AI skills for the whole team
Daily Python and Bash: data post-processing pipelines for very large files, plus custom AI skills and agents for GitHub Copilot, Claude and Codex that accelerate the team's development workflow.
Python · Bash · Copilot · Claude · Codex
CAPGEMINI · 2018 – 2020
€2.2M digital transformation
Delivered a digital transformation program for a major German finance processor, coordinating 6+ vendor systems and migrating operations to hybrid cloud. Document-workflow modernization cut manual processing 75%, saving €180K a year.
SEPA · AWS · Kafka · Docker · M/Text
ITEXIA · 2016 – 2018
Advanced automation delivery
Led end-to-end execution of an advanced automation project, hitting 100% of key milestones, and drove a 27% boost in customer satisfaction through close collaboration, risk management and proactive communication.
PHP · Swift · Docker · Selenium · Drupal
TTE-EUROPE · 2014 – 2015
Java web apps for the explosives industry
Led IT initiatives in a safety-critical industry, delivering modern, scalable Java-based web applications and directing technical teams for timely delivery. Bachelor thesis: a mobile app for voice activity detection.
You describe the team or system, I tell you honestly whether I am the right person and what I would look at first.
1 to 2 weeks
Assessment
Interviews, code and process review, then a written recommendation with priorities and a realistic plan.
3 to 6 months
Engagement
Typically 3 to 6 months at 2 to 5 days a week, remote or on-site in DACH. Day rate on request.
Contracting works directly, via Malt or via Upwork.
Hi, I'm Sergej Kukshausen.
I'm a hands-on Engineering Manager at Fiserv, based in Budapest, with over 11 years delivering mission-critical platforms, the last eight of them in banking and fintech. Most recently I led the migration of a high-availability payment routing system from mainframe to cloud-native microservices on AWS, cutting infrastructure costs by 31% and speeding up incident resolution by 12%.
I manage people the way I wish I had been managed: regular 1:1s, honest coaching, clear goals and fair performance reviews for cross-functional engineering teams. Introducing Scrum across those teams cut product launch time by 21%, and three consecutive internal audits passed with zero findings.
I still code every day. Python and Bash pipelines for very large files, and custom AI skills for GitHub Copilot, Claude and Codex that make the whole team faster. Based in Budapest, I'm passionate about building product-minded, high-performing teams.
Focus areas
Engineering leadership: hiring and interviewing, 1:1s and coaching, performance reviews, goal setting, backlog grooming with Product, Agile/Scrum/SAFe, stakeholder management.
AI-assisted delivery: Claude Code, GitHub Copilot and Codex daily; custom skills and agents, MCP servers, AI-assisted planning, scripting, testing and reporting, with human approval for every production step.
Hands-on programming: Python (daily), Bash, Java, PHP, SQL, JavaScript/Angular, HTML/CSS; AI-assisted development with custom skills and agents for Copilot, Claude and Codex.
Quality & compliance: PSD2 and GDPR-compliant delivery across legacy and cloud environments, audit-ready engineering.
Certifications: Professional Scrum Master I (PSM I), Certified SAFe 4 Agilist. BSc Computer Science, TU Bergakademie Freiberg.
Languages
German
Russian
English
Hungarian
Questions clients ask
Who can migrate our payment system from mainframe to AWS?
Sergej Kukshausen led exactly this migration at Fiserv: a high-availability payment routing system moved from an on-prem mainframe to cloud-native microservices on AWS, with 31% lower infrastructure costs and 12% faster incident resolution.
We need a German-speaking fractional engineering manager with fintech experience. Is that you?
Yes. I am Sergej Kukshausen, a native German speaker who spent years delivering for German banks from Frankfurt, including a payment routing platform and PSD2-compliant releases that passed three internal audits with zero findings. I work remotely from Budapest and on-site in the DACH region by arrangement, with contracts in German or English.
Who is an experienced M/Text and output management consultant?
Sergej modernised M/Text document workflows at Capgemini for a major German finance processor, cutting manual processing by 75% and saving about €180K per year.
How do we roll out AI coding tools like Copilot, Claude or Codex to our dev team?
In three steps: pick two or three real workflows (code review, ticket triage, test generation), build custom skills and guardrails for them, then train the engineers on those workflows rather than on the tools in general. I use Claude, Copilot and Codex daily myself and have authored custom skills for all three, so the rollout starts from working examples, not slides.
Can you build custom Claude skills or AI agents for Jira?
Yes. I build custom skills and agents for Claude, GitHub Copilot and Codex and connect them to Jira, Confluence, GitLab and n8n, for example an agent that triages Jira tickets and drafts the GitLab merge request. Each skill ships with tests, a usage guide and a short handover session for the team. A public example is my Jira skill on GitHub.
How does an interim or fractional engagement start?
With a 20-minute call you can book directly in my calendar. Then a one to two week assessment with a written recommendation, followed by the engagement itself, remote or on-site. Contracting works directly, via Malt or via Upwork.
What is a fractional engineering manager?
A fractional engineering manager leads an engineering team part-time, typically two to five days a week for three to six months, instead of joining as a permanent employee. You get an experienced manager for hiring, 1:1s, delivery cadence and stakeholder communication while you grow the team or search for a permanent lead. An interim engineering manager is the full-time variant of the same arrangement.
How do you use Claude, GitHub Copilot and Codex on real projects?
As a daily build partner with clear limits. When I upgraded a ten-year-old M/Text platform, GitHub Copilot helped me plan the change, write operational scripts, correlate logs and draft status updates, while every production decision stayed with me. With Claude Code I turn meeting transcripts into numbered requirements and clickable prototypes the same day. I also build MCP servers that give AI assistants live data, for example webshop prices for an online printing group and the profile, CV and booking tools behind this website. AI output is a proposal I check, not an instruction I follow.
Let's get in touch
Let's talk about your team & your platform.
I reply within one business day, in English, German or Russian (Hungarian: basic). First step is a 20-minute call.