Turris Next AI Language Audit

Services

All our services.

In detail, in one place.

We have organised what we offer into six areas: from AI language quality through translation and localisation to creative content. Choose what interests you, or scroll through.

AI language quality

AI should be accurate in every language.

We measure, test and continuously monitor how AI systems, chatbots and prompts perform in Hungarian and other languages — on your real content.

Illustration: a multilingual chatbot conversation with a magnifying glass and quality indicators

AI Language Audit

The aim of this service is to give the company a measurable picture, based on its own content, of how reliable the AI, machine translation or chatbot system it uses is in Hungarian and other selected languages. We don’t provide a generic technology rating. The assessment focuses on fidelity of meaning, terminology, style, consistency and serious language errors.

Multilingual AI agent / chatbot testing

Language testing of chatbots and agents examines whether the system answers real user questions clearly, with correct terminology, in a consistent tone and in the right language. The service focuses on the language quality of the user interface and the responses — it is not a technical security test, penetration test or legal compliance audit.

Linguistic and subject-matter evaluation of AI output

This hourly-rate service is a professional linguistic review of a given volume of AI-generated text. We evaluate accuracy of meaning, terminological correctness, readability, tone, consistency and fitness for the intended use. The assignment covers evaluation and error marking; a full rewrite or correction may be a separate item.

Linguistic optimisation of prompts and instructions

This service improves the linguistic, terminological and localisation quality of the prompts and system instructions given to the AI. The goal is for the AI to follow the brand voice more consistently, use the right expressions and reduce ambiguous instructions. It is not model development and does not guarantee error-free operation; we optimise a predefined set of prompts and check them against sample outputs.

AI-ready terminology and prohibited terms

We turn your existing corporate terminology into rules that both translators and AI systems can follow. The work can include preferred forms, terms to avoid, definitions, equivalents per language and usage examples. The recommended project assumes processing up to one hundred concepts supplied by the client in Excel; a larger database is quoted separately.

Localising brand voice and style rules

We turn the company’s tone of voice and writing rules into a guide that can be applied in each language. We define the form of address, sentence length, terminology, tone and cultural adaptation that suit the brand, then use examples to show the preferred solutions and those to avoid. The package covers one brand, a localised guide, ten examples and a review of twenty AI outputs.

Linguistic review of knowledge bases

A linguistic review of knowledge bases, help articles and customer support content improves clarity, terminological consistency and the usefulness of answers. This is especially important if an AI chatbot or internal copilot also draws on them. The package includes a linguistic review but does not cover RAG system development, source ranking or technical knowledge base architecture.

Regression language test set

A regression test set provides repeatable language examples for checking changes to a chatbot, AI agent or translation system. The starter package includes one hundred test cases, expected language criteria and a reusable dataset; regular test runs are agreed separately. This makes it visible if terminology or answer quality deteriorates after a prompt or model update.

Continuous AI QA and quality monitoring

As a monthly service, we sample your corporate AI outputs, check them against the agreed language and terminology rules, and flag changes in a short report. The basic package assumes one system, one language and up to one hundred samples of no more than one hundred words each per month. It is not a comprehensive guarantee covering every output; the sampling and thresholds are agreed in advance.

Have a question about this? The first consultation is free.

Let’s talk →

“An AI is only as good as the precision of its language. We work to make it mean the same thing in every language.”

— Turris Next

Translation

Translation, tailored to the risk.

From low-risk internal material to legal and medical texts: we always provide as much human control as the content requires.

Illustration: a joint translation workflow between AI and a human expert

AI + human translation Fast

For low-risk, internal or high-volume content, we use an AI draft, and a human checks an agreed sample. This is not full expert post-editing, so we don’t recommend it for critical content intended for publication. The sampling rate and acceptance criteria are agreed in writing.

AI + human translation Business

This is the recommended core package for business web, product and customer communication texts. After the AI draft, a professional translator fully checks and corrects the text, followed by language QA at the agreed quality level. The result is a human-verified translation ready for its intended purpose; the actual savings depend on the material, the language pair and the machine output.

MTPE of client-supplied machine translation

Machine translation post-editing is a full human correction of machine translation produced by the client. It is only cost-effective if the raw machine output is of adequate quality; if the machine text is flawed or misleading, retranslation may be needed. A preliminary sample review helps determine whether the job really is MTPE or full professional translation.

Critical AI + expert workflow

For legal, financial, medical, life-science or reputation-sensitive texts, the risk may justify expert retranslation or post-editing, an independent second review and documented QA. AI does not replace subject-matter accountability. We define the exact process once we know the document and the possible consequences.

Have a question about this? The first consultation is free.

Let’s talk →

“The machine is fast. The human is precise. A good workflow brings out the best in both.”

— Turris Next

Localisation and language tools

Localisation that works in your systems too.

Websites, software, online shops and translation memory: we prepare the content and the process so that languages flow smoothly.

Illustration: localising a website and mobile app into multiple languages

Website / software / app localisation setup

A technical preparation project that gets your content ready for localisation for one language and one export format. We set up the basics of text extraction and re-import, check the tags and run an import test. Translation is priced separately; actual connector development is not part of the package.

Product / PIM / e-commerce localisation setup

We align the language localisation of product data with your PIM field structure. We map the fields, mandatory data and term list, then verify the process in a fifty-product pilot. This helps prevent descriptions, attributes and categories from getting mixed up. PIM integration development and full product translation are separate items.

Continuous localisation coordination

Monthly-fee coordination for up to four release packages, one product and two languages. The service organises the handling of requests, files, statuses, approvals and language tools. Translation volume is priced separately, so the monthly fee does not cover unlimited content.

Multilingual SEO / market adaptation

An SEO and content plan for one market, including localised search intent, twenty keyword clusters and recommendations for five pages. The service is not just keyword translation: we validate search terms and target-market phrasing locally. Finished copywriting and technical SEO are separate assignments.

TM clean-up and AI-ready preparation

When cleaning up a translation memory, we remove or flag duplicate, incorrect, outdated or unreliable segments, then check a sample. The prepared TM can be used safely in translation and AI workflows. The service covers up to ten thousand segments and five hundred human checks; new translation is not included.

Have a question about this? The first consultation is free.

Let’s talk →

Have a question?

Let’s discuss what you need.

The first consultation is free. You tell us about your situation, and we show you which service or combination fits.

“Localisation isn’t the last step — it’s part of the product from day one.”

— Turris Next

Measurement and control

Measurable quality, risk under control.

We compare engines, check outputs and plan where humans are needed in the process — especially for confidential content.

Illustration: a quality measurement dashboard and a security shield

LLM / MT engine benchmarking

We compare up to three translation engines or language models on the same sample. The blind evaluation examines quality, terminology, errors, cost and speed. The project is based on up to one hundred segments and one language. Model API fees are a separate item; the results apply only to the content and settings tested.

Terminology / consistency / style QA

Hourly-rate quality checks of AI outputs that examine compliance with the approved terminology, style and consistency. Useful as an independent check if the content was produced by another team. Full correction, content rewriting or system configuration may be ordered separately.

Human review and approval plan

We plan at which points in a given workflow a human expert must check, correct or approve AI-generated content. We define roles, quality gates, escalation and acceptance criteria. This is a linguistic and operational plan, not legal or general AI compliance advice.

Language risk matrix and segmentation

We group content types by risk, volume, repetition, publication purpose and consequences of errors. On this basis, we recommend where automation is sufficient and where sampling, full post-editing or expert control is needed. The matrix covers up to ten content types and one workshop; it does not replace a legal risk assessment.

Requirements for confidential AI workflows

From a language workflow perspective, we define the data handling and tool usage requirements an AI task needs — for example, which content must be anonymised, which approved tools may be used and where human checks are required. We do not provide legal data protection or cybersecurity certification; final tool approval is given by the client’s legal/security team.

Have a question about this? The first consultation is free.

Let’s talk →

“What we don’t measure, we can’t improve. What is risky, we don’t leave to chance.”

— Turris Next

Strategy and operations

Strategy and lasting language operations.

We assess your current operations, set the direction for the next steps and, if needed, run your language operations month by month as your external team.

Illustration: a localisation strategy roadmap with milestones

Localisation maturity / workflow audit

Using up to three interviews, a process map and a priority list, we uncover where your current localisation is slow or error-prone. We examine the path of requests, terminology, translation memory, quality checks and technology connections. The audit provides recommendations; system implementation and software development are separate projects.

AI localisation strategy

A strategic project that recommends the right combination of AI and human control based on your target markets, content types and business risks. It includes a roll-out plan and business assumptions for up to three markets. The recommendation is technology-agnostic and based on the samples and client data available; we do not promise guaranteed savings.

TMS / CMS / PIM integration needs assessment

We assess how a content management, localisation or product data system could connect with one another. For one pair of systems, we prepare requirements, a content map, interface assumptions and a partner brief. This is an assessment — not software integration, custom connector development or vendor licence fees.

Managed Language Operations

As your external language coordination team, we handle requests, language tools, sample-based AI QA and reporting every month. The basic package includes twelve coordination hours, two languages and one workflow. Translation volume, software, licences and extra languages are billed separately. The monthly fee does not mean unlimited service.

Have a question about this? The first consultation is free.

Let’s talk →

“A good language strategy isn’t more translation — it’s a better decision about where a human is needed and where a machine is enough.”

— Turris Next

Creative content and media

Creative content, for every market.

Campaigns, copy, e-learning and video: we adapt the message so that it lands naturally in the target market too.

Illustration: video dubbing and creative content

Transcreation / campaign and slogan adaptation

Creative adaptation that conveys the intent and brand message of a campaign in a natural, culturally appropriate form in the target market. It is not literal translation. The basic package covers one market, a short brief, two versions and one round of revisions; additional concepts, legal review or research are separate work.

Multilingual copywriting

Target-language content written from a brief — either new copy or an AI-assisted first draft with expert review. The package covers one language, fact-checking and one round of revisions. We don’t price it by translation word count, because the time spent on research, ideation and approval matters more.

E-learning / video localisation engineering

Technical localisation engineering from an editable e-learning or video package. We extract the assets, handle the localisable elements, reinsert the text and carry out playback QA. Translation, narration and complex interactive development are priced separately; we bring in a partner where needed.

Dubbing project

A dubbing project for a short, simple corporate video with a single speaker, including script adaptation, a human voice and basic synchronisation. Pricing is per finished minute and per language. The final price depends on the studio, voice talent, audio length and post-production; AI voice and lip-sync are not automatically included in the package.

Have a question about this? The first consultation is free.

Let’s talk →

Next step

Not sure which one is right for you?

Briefly describe your situation, and at a free initial consultation we will choose the right service or combination together.

↓