When the Model Doesn’t Speak Your Language: Reliably Translating Low-Resource Languages

📅 Date: 11.03.2026

🎙️ Speaker: Rosen Krumov – Senior Machine Learning Engineer with expertise in multilingual NLP and low-resource language translation

🔐 About the event:
This event is part of the AI/ML/Big Data user group.

PROGRAM:
Large Language Models have significantly improved translation quality, but primarily for high-resource languages and generic text. In low-resource languages like Bulgarian, especially for technical materials, problems such as terminology drift, syntactic hallucinations, and domain misalignment arise. This talk explores designing robust, enterprise-grade technical translation systems for low-resource languages, using Bulgarian as a case study.

Presentation topics include:
>>> Failure patterns in Bulgarian technical translation
>>> Terminology grounding using knowledge retrieval
>>> Handling tables and CAD-linked terminology

Additional discussions cover:
>>> Incorrect case agreement in technical contexts
>>> Mistranslation of domain-specific terms

This session is aimed at ML engineers, applied researchers, and tech leads building multilingual AI systems beyond English-centric benchmarks.

>>> Q&A session included

The event is free but requires prior registration.

User Group: AI/ML/Big Data. Join the group for monthly events on AI/ML/Big Data topics and stay updated.

Organizer: DEV.BG – a specialized IT job board.

Job listings: Looking for AI/ML/Big Data jobs? Check listings on DEV.BG.

WHEN: 11.03.2026, 19:30

SPONSORS:
Iris.ai – https://dev.bg/company/iris-ai/
PROS – https://dev.bg/company/pros/

About the speaker:
Rosen is a Senior Machine Learning Engineer with over 10 years of experience building production-grade SaaS products. His background combines applied ML, system architecture, and enterprise AI deployment focusing on reliability, scalability, and maintainability. He specializes in deep translation, multilingual NLP, technical terminology, and domain-specific translation quality, bridging model capabilities with real product constraints for consistent performance in demanding environments.