There is a useful kind of knowledge that rarely makes it into a project page or a CV: the decisions that did not work, the trade-offs behind an architecture, and the small discoveries that change how you approach the next problem.

This blog is where I will keep those notes.

The threads I keep returning to

My work currently moves between software engineering, applied AI, sustainability, and product development. Those areas may look separate from a distance, but the problems underneath them are often remarkably similar.

How do we turn unstructured evidence into information that can be trusted? How do we make a complex decision traceable? How do we build useful AI systems under real constraints rather than ideal demo conditions? How do we learn from a product that was paused without treating the work as wasted?

I encounter versions of these questions while working on document intelligence for Pardus AI, researching multimodal greenwashing detection, building sustainability and formulation products, and maintaining open-source tooling around AI infrastructure.

What I plan to write

The writing here will generally fall into four connected areas:

  • AI and document intelligence: retrieval, multilingual embeddings, reranking, document transformation, evaluation, and the difference between a convincing prototype and a dependable workflow.
  • Sustainability data: reporting standards, evidence trails, emissions data, greenwashing signals, and why auditability matters as much as automation.
  • Product engineering: architecture choices, change-impact thinking, security, interoperability, and lessons from taking products from an idea to a working system.
  • Field notes: honest retrospectives from active, completed, and paused projects.

The goal is not to manufacture certainty. I would rather document what I tested, what I observed, what remains unclear, and how the evidence changed my mind.

One source, several conversations

Long-form posts will live here first as the canonical version. LinkedIn will carry shorter, conversation-oriented adaptations, while selected articles may also be republished on Medium with a canonical link back to this site.

That structure keeps the archive independent without pretending that distribution does not matter. The website remains the source of truth; other platforms become places where each idea can meet a different audience.

This first note is simply a marker. The useful part begins with the next problem worth explaining.