Skip to main content

# JOURNAL

long-form notes from the build.

Engineering posts only — opinions, lessons, and technical deep-dives from two people who run data systems in production for a living.

subscribe via RSS

  1. #ai

    The EU AI Act takes effect August 2, 2026. Here's what mid-market DACH companies actually need to ship.

    The EU AI Act takes effect August 2, 2026, with penalties up to 7% of global turnover. Five practical steps mid-market DACH companies actually need to ship.

    Juri Saloid · 2026-04-17 · 8 min

  2. #data-engineering

    Client-side GA4 is leaking 30% of your data. Here's the fix.

    Client-side GA4 leaks 25-40% of pageviews to ad blockers and DACH consent rejections. Server-side tracking closes the gap — and improves GDPR posture.

    Juri Saloid · 2026-04-17 · 8 min

  3. #ai

    Why AI pilots fail before production — and the 90-day fix

    Over 70% of enterprise AI pilots never reach production. Here are the 5 reasons they stall — and the 90-day framework that actually ships them.

    Juri & Oleks Saloid · 2026-04-16 · 8 min

# WHAT-WE-WRITE

engineering notes from production.

This journal covers the systems we actually ship for mid-market companies in DACH — server-side tracking architectures, AI and RAG systems in production, EU AI Act readiness, EU-sovereign cloud platforms, and big data pipelines on AWS, GCP, and bare-metal EU infrastructure. Not opinion pieces or industry takes; engineering posts where we work through a real problem end-to-end and document what worked, what didn't, and the operational discipline behind it.

We don't run a content calendar. Posts ship when a lesson is durable enough to be worth your reading time — usually after we've solved the same problem more than once for different clients and a pattern is stable. The format favours depth over breadth: long-form deep-dives, concrete numbers, and the trade-offs we made on real engagements.

A few recurring threads run through the archive. The data-loss math behind client-side GA4 in DACH (typically 25-40% of pageviews missing on a standard setup) and what server-side tracking with GTM Server-Side and proper consent management actually recovers. MLOps and observability for LLM-backed systems — model evaluation, drift detection, agent architectures, and the discipline that separates a Jupyter notebook from a production AI agent that runs reliably without you in the room. The EU AI Act compliance picture for mid-market deployers under the August 2026 deadline: what's actually on the technical-controls list, what's optional, and how to evidence it during an audit.

We also write about EU-sovereign architecture — Hetzner and bare-metal EU hosting choices, BigQuery and Snowflake alternatives that keep data residency clean, and the practical Terraform and Kubernetes patterns we use to run production data systems on infrastructure that holds up legally and operationally. And the unglamorous parts: data quality in production, deployment automation, monitoring that pages the right person, and the cost trade-offs of distributed processing with Spark and Hadoop versus simpler architectures.

If you want new posts in your reader instead of checking back, the RSS feed is linked at the top of this page. For consulting on any of these topics for a DACH mid-market company, the contact form on this site goes straight to us — there are no account managers in the middle.

want this in your inbox?

one short note per month. only when we have something worth reading.

 subscribe-via-rss

// or write: hello@saloid.com · gräfelfing · de