AI-first by design. Compliance-first by architecture.

AI-first, what it actually means

When we say AI-first, we mean the AI is the substrate, not a feature bolted on. Two feature pages show what that looks like where you work every day.

The content pipeline is the upload story. You upload a video; the platform transcribes it, translates it into the languages you've configured, generates subtitles for each, prices the content for every market you sell into, and files everything into your reusable catalog. The work is the upload. Everything downstream happens automatically, on every video, forever.

Semantic search is the discovery story. Every lesson, transcript, PDF, slide deck, community post and live recording is indexed as one searchable surface. A member asks in natural language and gets the exact moment that answers the question — across every course they have access to and every thread they're a member of. The archive stops being a place things go to die.

These aren't two products that happen to share a sales page. They're two views into one AI surface — the same content catalog, the same audit trail, the same EU-resident inference.

That last part matters for the rest of this page.

The same AI runs the selling

The back-office work on what we handle — tax applied at checkout, fraud screening, failed-payment recovery, invoicing, disputes — is the part of selling nobody wants back. It's also work that has to keep up with rules that change every year.

Most platforms run that on static rule tables that engineering teams update by hand — which is why new requirements take quarters to land elsewhere, and why smaller markets often never get supported at all.

On lernaura, the operational decisions — the price a buyer in Stockholm sees in kronor, the fraud signal that catches card-testing at 3am, the retry timed to actually recover a failed renewal — run through the same documented AI surface as everything else. Every decision is logged and traceable, line by line.

What that means for you is simple: the back office keeps up with the rules without you noticing, and when anyone asks how a decision was made, there's a written answer.

Today that covers price localisation, fraud screening, payment-recovery timing and the language work behind your documents — and the coverage keeps widening, with each new capability documented the same way from day one.

It's also why the compliance posture has to be as serious as the AI posture. Here's what that looks like.

Compliance-first, what it actually means

A compliance posture is either part of the architecture or it isn't. There is no third option that gets discovered later.

The EU AI Act requires AI providers and deployers to document their systems: which model, what data, what purpose, what risk. Platforms that wrapped LLM calls into features in 2023 cannot produce those records after the fact. On lernaura, the documentation is what the platform produces as a side effect of running — which is why we can hand it over instead of promising it.

The mechanism is versioned skills.

Every AI action on the platform runs as a skill: a named, versioned unit with declared inputs and outputs, a specific model and version, an explicit purpose, and its EU AI Act and GDPR Article 22 classification decided up front. Transcribing your video, pricing for a Swedish buyer, ranking a search result — each is a skill, and each run leaves a permanent record.

Four things fall out of that for you:

• Every AI call is traceable. Which skill, which model and version, what went in, what came out, when, and for whom — logged permanently. When a question comes six months later, the answer isn't a reconstruction; it's a lookup.

• The paperwork already exists. When a regulator or a procurement reviewer asks "show us how this AI works", the answer is the skill catalogue and the logs — available on request, not assembled for the occasion.

• Decisions about people stay human. Skills that could touch grading, certification, access or behavioural inference are restricted to advisory-only mode with a human in the loop. The mechanical ones — transcription, translation, search ranking, price localisation — are documented as exactly that, with the reasoning preserved.

• Model changes don't erase history. When a model is upgraded, that's a new skill version; everything that ran under the old one stays attributable to it. "Which model produced this output in March?" always has an answer.

This is what compliance-first means here. Not "we're working on AI Act compliance" — but that the documentation the Act asks for is produced every time the platform runs.

What it looks like in practice

A concrete example. A creator in Munich uploads a 45-minute lecture. That single upload leaves, in the audit trail:

• Transcription — input: the audio; output: a transcript with timestamps
• Language detection — input: a transcript sample; output: de-DE
• Translation — input: the German transcript and the configured target languages; output: a transcript in each
• Subtitle generation — input: the timestamps and translations; output: subtitle tracks in each language
• Search indexing — input: the transcripts; output: embeddings, stored in the EU
• Price localisation — input: the EUR reference price; output: prices for each market the creator sells into

Six entries, each permanent. If a licensee's DPO asks six months later which AI processed their team's training content, the answer is the log. If a regulator asks whether the transcription makes automated decisions about people, the answer is the skill definition declaring it doesn't — plus the evidence behind it. The same applies to every search a member runs and every check on every transaction.

For creators and sellers, this means the AI Act obligations that would land on you as a deployer are dramatically reduced. We do the upstream documentation work; you inherit it.

For B2B buyers, it means your AI-governance lead gets written answers to the questions 2026 procurement reviews now ask — produced continuously, not assembled for the deal.

What sits behind this

An EU-resident substrate, end to end: the models, the inference and your data all stay in Europe, with contractual no-training terms on your content. The full subprocessor list is on the European platform page. The DPA is embedded in our Terms of Use, accepted at signup, the same for every customer including the free tier.

SOC 2 Type II and ISO 27001 are both in preparation. Because the platform was built this way from day one, that work is formalising controls the architecture already implements — not retrofitting them.

Try lernaura

The point of building AI-first on an EU substrate, with the compliance produced by the architecture, is that your next conversation gets easier — with a buyer, a DPO, or a reviewer who has been told "we'll get you the documentation" everywhere else.