Insights on digital experience | QED42

All blogs

All-five accuracy on hard messages by training size (log scale): 31 fine-tuned Laya runs on Kaggle T4 x2 (4 Oct 2026) against TF-IDF on the same seeded subsets. Bars show the worst to best of five runs. Dashed lines are the zero-label systems.

Jev and Laya, tested on our own data: what held up

An independent experiment testing Jev and Laya, two new typed decision models, on a year of our own Slack availability messages: zero-shot accuracy, fine-tuning, calibration, cost and latency.

Laptop connected to a person icon on orange background, illustrating a network link.

The digital sovereignty questionnaire

This is a ten-minute, 16-question self-check that scores how much real control your organization has over its data and its AI-enabled workflows. It runs entirely in your browser and saves nothing anywhere. If you would rather just take i...

Laptop icon with sleeping "z" symbols in speech bubble, on brown background.

Digital sovereignty: Knowing what must stay under your control

In a nutshellDigital sovereignty is becoming a more important enterprise concern because the systems businesses rely on are no longer passive infrastructure. They store data, shape workflows, support customer-facing services, connect tea...

Code intelligence built around how Drupal actually works

Code intelligence built around how Drupal actually works

IntroductionMost scanners read PHP and stop there. EventHorizon reads the hooks, services, config and caching layers that decide whether a Drupal platform is healthy, and turns the discovery phase of an engagement into a visual, defensib...

HAAT deep dive: turning human intervention into a scalable primitive

HAAT deep dive: turning human intervention into a scalable primitive

Unlocking high-latency authority in autonomous agentsThe "Babysitter" problemIn standard "Human-in-the-Loop" (HIL) implementations, the human is treated as a supervisor.‍The agent runs until it fails.The system pauses or crashes.A develo...

Responsible AI testing: ensuring ethical, fair, and safe AI systems

Responsible AI testing: ensuring ethical, fair, and safe AI systems

These days, wherever we go, be it a tech meetup, Conference, or our daily conversations, we hear the same word again and again, i.e., AI. ‍AI is now used extensively in almost everything; all our apps, devices, and everything has AI...

Building a video RAG system that's 81% cheaper than "Industry standard", here's how

Building a video RAG system that's 81% cheaper than "Industry standard", here's how

The $2,700/month reality checkLet's talk about what it actually costs to build Video RAG in 2026. Standard approach everybody recommends: sample frames at 1 FPS, send every frame to GPT-4o Vision ($2.50/M input, $10/M output tokens), gen...

Create your own AI agents for Drupal Canvas

Create your own AI agents for Drupal Canvas

Drupal Canvas has officially launched, and many of you are likely already building impressive projects with it. As you may know, Drupal Canvas ships with several "out-of-the-box" AI features, such as AI-assisted code component generation...

Building simple & effective prompt-based Guardrails

Building simple & effective prompt-based Guardrails

Why AI systems need GuardrailsThe wild west days of early AIWhen ChatGPT first captured public attention, it didn’t take long for users to start testing its limits. Among the earliest and most notorious exploits was "DAN" (Do Anything No...