KATE LOGINOVAPERSONAL / CURIOUS
Business · Signal & Sense

Projects, past and present

Things I helped build.

Some research, some engineering, some products that eventually had to survive real users. None of this happens alone — here's the part I played.

For services or a commercial collaboration, contact Signal & Sense.Signal & Sense ↓
Production AI

Clinical NLP that works in production.

At Dedalus, I worked on multilingual clinical language pipelines — extracting entities, negations and relationships from demanding healthcare text.

NLP / MLHealthcareSystems architectureEvaluation

The challenge

Clinical language is inconsistent, specialised and consequential. Useful extraction requires both modelling quality and reliable engineering under real operational constraints.

My contribution

  • Developed and improved multilingual extraction components.
  • Refactored evaluation and model-development workflows for reproducibility.
  • Worked on model serving and throughput improvements, including Triton migration.
  • Supported cross-team validation and knowledge sharing.

Selected result: up to 8× processing-speed improvements on individual components.

Public-interest technology

From information noise to testable questions.

Through Aletheos and collaborations with journalists and civic organisations, I work on tools that help people examine claims, compare coverage and understand the information environment.

Information integrityMedia analysisLLM evaluationCivic technology

The challenge

Information harms are not solved by a single classifier. Journalists and researchers need traceable evidence, realistic workflows and careful treatment of uncertainty.

My contribution

  • Designed and developed AI-assisted fact-checking and media-analysis workflows.
  • Worked on manipulation detection and cross-platform monitoring research.
  • Connected technical exploration with the questions and constraints of civic users.
Teplitsa ↗
Applied language technology

When finding information is the hard part.

At Faktion, I worked on semantic search and language tools for complex documents, including Dutch government materials and English legal text.

Semantic searchInformation retrievalText quality

The challenge

People do not ask questions in the same language that documents use. Search systems must bridge vocabulary, context and intent without losing reliability.

My contribution

  • Research and development for semantic search and classification.
  • Query expansion, correction and word-sense disambiguation.
  • Prototyping, API work and user feedback integration.
  • Research on readability, fluency and coherence.
Research & prototyping

How do machines answer questions across languages?

At DFKI, I researched multilingual non-factoid question answering, building neural methods and an interactive demonstration of attention mechanisms.

Multilingual QADeep learningResearch communication

The challenge

Answering open-ended questions is not simple fact retrieval. It asks models to represent relevance, context and meaningful linguistic variation.

My contribution

  • Developed recurrent and convolutional neural approaches.
  • Built a visual demonstration to communicate model behaviour.
  • Presented work in research venues and taught related material.
Explore the research code ↗
Where I've worked

A few places
I've learned.

Started 2022

Dedalus

Clinical NLP and applied ML engineering, moving research components toward production.

2021–2022

Faktion

AI consultancy: retrieval, semantic search and text intelligence.

2018–2022

Ghent University

PhD research in NLP for personalised education.

2017–2018

DFKI

Research on multilingual question answering.

Personal and professional

Personal projects here. Business through Signal & Sense.

This is my personal archive. AI consulting, product work and commercial collaborations go through Signal & Sense. Research, teaching, books or just saying hello? Use the personal email below.

Business contact ↓