Currently working on an agent-based app family for making recipes easier to manage.

The main insight of the app -> use LLMs to turn unstructured recipes from multimodal sources (images, pdfs, websites, videos) into structured data that can be easily searched, organized, and used for meal planning and shopping.

Some core problems I want to solve:

• Super easy recipe collation from multiple sources (images, websites, pdfs, or videos)
• Agent-assisted meal planning
• Smart shopping list creation from meal plans -> using agents to combine ingredients and quantities and organize them by grocery section (e.g. produce, dairy...)

The possibilities are endless and I'm motivated by solving real problems that I face as a father and home cook.

How I'm doing it (so far)

• .NET for the core data model and related MCP tools (e.g. creating recipes, getting ingredients, etc.)
• Next.js for a web app to rapidly prototype real features and flow.
• Mastra.ai for agent orchestration and graph-based workflows
• k3s for production-ready (but not overkill) infrastructure and platform.
• Github and Actions for source control and CI/CD
• Postgres as my DB

However things are changing rapidly. 🚀

After the core functionality and user management are solid I will focus on an ios app and package the web app as a PWA and/or an installable desktop app. ; I'll be blogging about my progress and learnings as I'm learning a lot about how to make actually valuable and cost-effective agent-powered software.

In 2020 I moved to Germany and joined the Planner team at Heidelberg Materials. Check my Tech Lead experience for more details, but Planner filled a gap in the cement planning process - namely factoring in electricity price volatility. By doing so we were able to generate massive savings for the company.

Keep in mind this was mid-covid so I really got a chance to learn about remote work with an international team.

The work we were doing was unique and really valuable. We didn't just save money, but we aligned the production of cement to use renewables as much as possible. At the scale of Heidelberg Materials any optimization is a huge saving in CO2 emissions worldwide.

Notable Achievements

The project started as a POC, and I entered and led the UI development. Fun fact, our UI later won the UX Design Award in 2023.

I rebuilt most of the early POC into a scalable foundation for the future:

What I did in a nutshell was completely redo most of the initial hasty POC implementation and set up a solid foundation for the future of the product. This included:

• Refactored state management → removed Redux, switched to React Query
• Offloaded heavy computations to Web Workers to unblock the main thread
• Developed high-performance charts, sometimes writing SVG directly for specific requirements
• Introduced modern UX patterns: skeleton loaders, bundle splitting, link prefetching, optimistic updates

Cross-Team Impact

Because we worked in a monorepo, many of my changes benefited other teams too:

• Shared logging library used by all FE teams
• Unified CI scheme adopted company-wide
• Migration from Sentry → Azure Application Insights
• Created reusable UI components like an accessible multi-thumb slider

I also contributed to backend development, implementing a new scheduling API that enabled multiplayer planning for our users in the cement plants.

I was pretty proud of setting up the FE for long-term maintainability and scalability - especially when it comes to performance.

If I could do it again...

1. Push harder for Tailwind from day one (back then it was a tough sell!)
2. Define routing strategy earlier to avoid late refactors
3. Move from D3 → Chart.js sooner — D3 was overkill for most needs
4. Establish a design system early to ensure reusability and consistency

I led a cross-functional team of four engineers supporting and developing the organization’s ERP (Enterprise Resource Planning) and EAM (Enterprise Asset Management) systems, which were critical to 24/7 operations across seven departments.

To manage the growing volume and complexity of requests, I established a cross-departmental working group that aligned priorities, brought internal technical needs into the roadmap, and improved overall visibility and collaboration.

Responsibilities & Achievements

Under my leadership the team and I achieved significant improvements:

• Delivered major platform upgrades — from VM-level improvements to full system migrations and version upgrades.
• Introduced modern DevOps tooling including Octopus Deploy and Jenkins for automated deployments.
• Implemented version control and CI/CD pipelines for ERP development — a non-trivial challenge due to legacy constraints.
• Created staging environments with automated production data mirroring and anonymization.
• Built test environments with automated data restoration.
• Hardened infrastructure following CIS benchmarks and best security practices.
• Standardized JIRA workflows and introduced Kanban methodology to reduce chaos and improve delivery predictability.

These changes transformed our work from reactive firefighting to proactive, faster feature delivery and reduced system downtime.

Individual Contributions

• Developed a vacation request application in .NET MVC, integrating directly with the HR ERP system.
• Built an event management tool for large-scale event registration and attendance tracking.
• Created an inventory and asset management system (with desktop client) for telecom equipment, including custom billing and partner integrations.
• Onboarded the purchasing department into the EAM system, establishing workflows for purchase orders and inventory management.

Knowledge Sharing & Documentation

I authored extensive internal documentation and training materials in collaboration with each department to mitigate knowledge loss from turnover.
To systematize this, I introduced the Divio documentation system and processes across the IT department, establishing a sustainable framework for maintaining up-to-date, high-quality documentation.

My first experience in a company, I got a chance to work on a very interesting project for Axciom, a marketing solutions company.

This project would sell models trained to calculate consumer propensity to buy different products based on the massive amount of consumer data the company collected.

It was written in Java, and I had a chance to work with the core product, getting my first introduction to patterns like Dependency Injection and Inversion of Control with the Guice framework.

I also got to work with a lovely team of supportive developers, from the local team to also some very experienced remote developers in the Czech Republic.

I'm still very grateful for this initial introduction to the industry, even though I have some moral issues with the scale of data collection and marketing involved.