EU-based AI data team · Croatia
Annotation, LLM evaluation and RLHF by native speakers, under NDA and GDPR, with built-in quality control.
Services
Rating and ranking model responses, writing reference answers, preference data and safety review in Croatian and regional languages.
Mathematics problems and step-by-step solution review by holders of master's degrees in mathematics.
Crop, weed and plant-disease labelling on field and aerial images, reviewed by an agronomist.
Classification, sentiment, named entities, translation and localisation review for regional-language datasets.
Bounding boxes, polygons and segmentation in your tool or ours (Label Studio, CVAT).
Transcription, audio labelling and native-speaker recordings across regional accents.
For companies running AI in Croatian
Your chatbot or AI assistant talks to customers in Croatian every day. Do you know how often it gets the facts wrong, sounds unnatural or gives risky advice? Our native-speaker team reviews a sample of real (anonymised) conversations and gives you a clear report with numbers and concrete fixes.
Request an audit01
We review 500–1,000 conversations and flag factual errors, wrong or unnatural Croatian, unhelpful answers and unsafe advice. You get a report with error rates by category and examples.
02
Every week we review a sample of conversations plus the cases where your AI was uncertain, and send back corrections you can use to improve the model.
03
EU-based, data processing agreement (GDPR) and NDA for every engagement. We work only with anonymised data inside your tools.
Sample output
Prompt
Koliko dugo se kuha grah?
Response A
Suhi grah namočite preko noći, a zatim ga kuhajte 1 do 1,5 sat dok ne omekša.
Response B
Grah nije potrebno namakati, kuhajte ga 15 minuta i gotov je.
A gives the soaking step and a realistic cooking time, while B's 15 minutes would leave dry beans undercooked.
Tight bounding boxes, occlusion flags, 10–20% second review.
Sample annotations on stock photos (Pexels), not client data.
Why Tehneo
Croatian, Serbian and Bosnian from native speakers who understand dialects, culture and context.
Master's degrees in mathematics and agronomy, plus hands-on experience in construction, hospitality and automotive services.
Your data stays in the EU, with an NDA and a data processing agreement for every project.
Guidelines, pre-labelling and QA designed for machine-learning use, so the data fits how models are trained.
How we work
01
Languages, data type, volume, tool, timeline and quality targets.
02
Free 50-item pilot so you can judge our quality before committing.
03
Trained annotators work to your guidelines, with weekly progress updates.
04
Gold-set checks, 10–20% second review and agreement tracking on every batch.
Security
Contact
Tell us about your project. We reply within one business day and can start with a small pilot.
Marko Balikić · Tehneo d.o.o. · Višnjevac (Osijek), Croatia
