AI & Generative AI

We turn current AI technology into something that runs inside real workflows and services.

LLM-based systems
RAG
Natural language processing
AI data analysis
Vision AI
Generative AI
Workflow automation

Adopting AI is not about picking a model. It starts with finding the point in a workflow where AI actually saves time. MEDIANAVI led the development of a news-domain language model and built the analysis logic that turns raw sensor readings into real inventory figures. Our job is to make it a system that runs every day, not a paper.

What we deliver

LLM-based work systems

Search, summarisation, classification and drafting — the repetitive text work people do by hand — handled by an LLM. We add a RAG layer so answers are grounded in your own documents, and show the sources so results can be checked.

Domain-specific language models

Models that reflect the vocabulary and writing style a general-purpose model handles poorly. We led the development of KPF-BERT, a news-domain language model for the Korea Press Foundation, and built article summarisation and clustering on top of it.

Vision AI & sensor analysis

Pulling meaning out of data that means nothing in raw form — LiDAR point clouds, images, time-series sensor readings. Reconstructing the inside of a grain silo in 3D to compute the actual stored volume is one example.

AI training data preparation

Collecting, cleaning, labelling and reviewing raw data until it can be used for training. We are a designated AI-processing supplier under the Korea Data Agency data voucher programme.

How we work

We decide which step of which task is being replaced, then prove that one step with a PoC.
We confirm the accuracy target against real data before scaling it into a system.
We design the review path so a person can check and undo what the AI produced.
Models change, so we keep the architecture able to swap the model out.