Everything below actually runs — built in our own lab, on our own infrastructure. For every showcase we lay the technology bare; you see the result live with a single click.
The AI assistant in the bottom right of this site. Hybrid retrieval over 812 fragments of our own pages, a source under every answer, and a threshold that does not even call the language model when nothing is found. 20 out of 20 on the golden set.
3.6 million rulings made searchable with RAG that backs every claim with a source. FTS5 retrieval in milliseconds, citation verification, on-prem.
Seven specialised agents that together find leads, diagnose sites, build demos and pitch them — with cost and quality monitoring at every step.
Measure how readable your site is for AI assistants: 15+ checks in parallel, AI content analysis and a roadmap that provably works (case: 36 → 85+). The second entry point of the website project: for those who already have a site.
Two agency sites and 26 complete demo sites, each live on its own domain. An image pipeline that turns two keyframes into a scrollable video, and a code chain that keeps the whole frame sequence under 10 MB — with a link crawl of 1,273 URLs and not one broken reference.
An AI NOC with guardrails: the brain may only propose, the hands may only act within guardrails, and everything is recorded in a single append-only log per incident.
Malware scans, file integrity and suspicious connections in a single security suite — with quarantine instead of instant deletion and one admin panel for scans, recovery and whitelisting.
A Dutch AI news site that writes itself: RSS clustering decides what counts as news, an AI editorial team writes with sources, the edge publishes. 4× a day.
An autonomous, learning trading agent that turns news and market data into substantiated positions — paper-only, with strict risk guardrails and every trade traceable back to its source.
One router that picks the cheapest suitable model per task, actually forwards the call and records per call what happened: route, real upstream, tokens, pricing status and savings. Whatever cannot be priced stays empty instead of zero — with ten screens from the running app.
AI code review on every pull request: it verifies the webhook signature, routes the diff to the cheapest model that fits, and posts back a single substantiated review — free for open source, with the cost of each review recorded.
Around 94 containers across two hosts, with a manifest that records what is supposed to be running. Rebuilt after a real crash: critical services deliberately run outside the stack they monitor.
Publicly reachable without a single open port on the router: everything runs over an outbound tunnel, with Workers at the edge, Zero Trust in front of the admin interfaces, and DNS managed as code.
210 workflows that drive the lab — plus two honest lessons about automation we deliberately rolled back because it did more harm than good.
A supervisor that reads backup runs, flags anomalies and warns when disks fill up — but never deletes anything itself. Data is sacred.
From parked domain to a living astronomy platform with an autonomous AI engine that measures space weather, reviews its own work and writes its own daily journal.
Every showcase is the same approach applied to a different domain: retrieval first, sources mandatory, autonomous where possible, guardrails where necessary. Tell us about your data.
Digital Salesperson
Neuralex
You are talking to an AI assistant. Every answer shows its source.
You are talking to an AI assistant of Neuralex — not a human. I answer only from what is on this site, with the source below each answer. If it is not here, I say so.
No legal advice · nothing stored in your browser