Data Scientist & LLM Engineer

UUG.AI turns camera streams into useful, real-time signals for operations teams. As a Data Scientist & LLM Engineer, you turn recordings, detections, tracks, events and operational context into dependable product features. Build evaluated, observable language and vision-language workflows that help users understand what happened.

Your responsibilities:

• Build vision-language and LLM workflows for summarisation, event understanding, search and operational automation.

• Design evaluations, test sets and feedback loops measuring quality, safety, latency and cost.

• Explore event and media metadata to identify patterns that can become product capabilities.

• Develop retrieval and context strategies across structured data, model results and selected media evidence.

• Productionise experiments as tested services or workflow stages, with monitoring and clear failure behaviour.

• Collaborate with product and engineering colleagues to scope problems and iterate from feedback.

Strong Python skills and the ability to analyse data and build production software. Practical experience with LLMs, vision-language models, retrieval or adjacent applied-ML systems. Ability to evaluate probabilistic systems beyond hand-picked examples. Comfortable with structured and unstructured data, provenance, access and data quality. Clear communication of trade-offs and a preference for simple, testable approaches.

Helpful, not all required: RAG, tool use, agent harnesses and structured output; multimodal text/image/video models; MongoDB aggregation, vector search or large event datasets; evaluation tooling, observability and prompt versioning; media pipelines, computer vision or time-series analytics. Relevant skills and practical experience matter; no mandatory diploma or fixed number of years.

Full-time position with hybrid collaboration from Ghent. Work on real product capabilities across video data, language models and multimodal AI. Collaborate with product and engineering colleagues, establish meaningful evaluation baselines and grow end-to-end ownership from experiment to production and continued improvement.