Health and Work AI Lab projects

Key technologies: Web mining, Natural Language Processing, Anomaly detection,

An AI‑Driven Pulse Monitoring System is a text‑mining tool that scans large volumes of public online content—such as parliamentary debate transcripts, thousands of million tweets, and news articles—to track how important societal and political topics evolve over time. Using advanced AI pipelines, the system identifies patterns, shifts in public debate, and changes in mood, giving a detailed and up‑to‑date picture of what people and politicians are talking about. This high‑resolution view helps detect early signals of changing public sentiment and offers a scalable way to monitor conversations across multiple sources.

Technologies used: RAG, LLM.

LLM-based chatbot for parents with questions about upbringing of their child, based on guidelines and education. Guardrails are incorporated for escalation to professional.

Key technologies: Simulation.

The Carebots project is a pilot exploring how autonomous robots can support hospital logistics. The focus is on helping staff—especially pharmacy assistants, who experience high workload—by automating routes where bulky or time‑sensitive items must be transported. The pilot studies robot behavior in real hospital conditions, including navigation among people, handling manual doors, timing deliveries, and managing risks like lift usage or dependence on staff. It also identifies opportunities for improvement, such as larger drawers, better flexibility, shared control, and reducing anxiety for children.

https://www.nlrobotics.nl/carebots

Technologies used: Ontologies, RAG, LLM.

LLM-bases system to provide decision support based on care protocols and unstructured data in the EHR.

Key technologies: AI, body keypoint tracking, anomaly detection, multi-modal fusion, spatio-temporal graph-based machine learning, privacy enhancing technology.

Subtle patterns or abnormalities in movement, speech, and language behaviour often go unnoticed through traditional assessment methods, even though they may represent early signs of disease onset. As a result, diagnoses are frequently delayed, and opportunities for timely intervention and treatment are missed. Artificial intelligence (AI) offers the potential to support healthcare professionals by improving the accuracy and efficiency of early detection through a multimodal approach. Our team is developing and evaluating the feasibility and design of a reliable, privacy‑preserving, AI‑driven system that integrates video, speech, and additional parameters to help identify early indicators of Duchenne Muscular Dystrophy and other developmental disorders in children. Early diagnosis can substantially reduce both healthcare costs and human suffering, making this approach highly promising from both clinical and societal perspectives.

https://openlab.tno.nl/projecten/ai-ontwikkelings-aandoeningen-kinderen/

Technologies used: LLM, Data Analysis, Data Vizualization.

A data-driven dashboard that uses LinkedIn data from RevelioLabs to explore labour market dynamics across Europe. The tool enables cross-country comparisons of skills and career transitions, provides information on career trajectories, including skills and occupations. It also provides insights into geographical distribution of occupations.

The dashboard currently covers six countries - the Netherlands, Italy, Poland, Bulgaria, Germany, and Luxembourg – and contains data from over 26 million unique job profiles.

It also includes a 2D representation of the embedding space of all of the ESCO skills and knowledges. Allowing the user to analyze skills and knowledges distributions for specific occupations. This tool might serve the researchers and legislators with interest in the labour market landscape in Europe.

https://diamonds.tno.nl/european-labour-market-dashboard/external-app/run/21

Technologies used: LLM, NLP, self-training AI models.

INoVA Hub provides a smarter way to explore screening models. Its main purpose is to support researchers in the quest for suitable screening models to test their hypotheses. This includes non-animal innovations like organoids and in vitro systems, but also the animal-based models in use.

https://inovahub.tno.nl/

Key technologies: Speech-to-text, LLM.

As experienced maintenance and installation workers age and retire, their extensive knowledge retires with them. Transferring that knowledge is the key to long-term business success, but creating accurate, effective instructions can be challenging and costly. We have developed a tool that automates this process, reducing the barriers to creating work instructions and thereby increasing productivity. With Instant Instructions, workers use a tablet to film each other performing routine and specialised maintenance and installation tasks, and explain each step of what they are doing. The power of AI then translates that input into easy-to-follow, step-by-step written instructions and accompanying video of each step. Once uploaded into the platform, the instructions can be controlled for quality, improved where needed, and translated into any language that might be required.

https://ventures.tno.nl/portfolio/instant-instructions/

Award-winning automated text coding system, winning the Eurostat "Occupations for Online Job Advertisements Challenge" and evolving into our ObjectivEye spinout.

Key technologies: LLM, RAG, Classification.

https://statistics-awards.eu/announcements/winners-wi-2nd-round

KnowledgeMiner is an online AI-based tool that assists experts in systematically finding evidence in scientific literature. It does so by facilitating the creation and expansion of dedicated ontologies and using them to mine substantial volumes of literature. Results are presented in heatmaps and datatables that can easily be used to further explore, analyze and filter relevant results.

https://diamonds.tno.nl/projects/knowledgeminer

Key technologies: LLM, speech-to-text, text-to-speech.

The personal health chatbot (PHCB), allows a person and their wellness coach to gain a 360-degree view on the person’s current wellness supported by a unique 360-degree visualization. Rather than having to fill in lengthy questionnaires, the person can simply chat (either by text or voice) with the PHCB to gather information and build up a wellness profile over time. This chatbot utilizes diverse AI components including different guardrails and could be adapted to other projects utilizing questionnaires.

Key technologies: Natural Language Processing, Anomaly Detection, GenAI.

Safety Stethoscope is an AI‑based early warning system that continuously monitors many operational and human‑factor signals—such as sensor deviations, maintenance workload, shift‑report sentiment, incident history, and workforce conditions—and combines them into a single, easy‑to‑read safety status. By analysing patterns in this data, the AI detects weak signals that may indicate rising safety risks and presents the outcome through a simple meter that helps operators spot irregularities early. It is always available, remembers past context, and supports people in making quicker and better safety decisions without replacing their judgement.

Technologies used: Text-extraction, LLM.

The SDS Entity Miner contains a pipeline using large language models (LLMs) to automate the extraction and management of data from SDSs to online chemical inventories. The pipeline achieved an average accuracy of 0.83 in (close to precisely) extracting multiple variables of interest, such as company name, product name, and hazard statements, in comparison to manually extracting these variables. Overall, this pipeline illustrates the ability of LLM tools to automate SDS inventory management and thereby support the possibility to perform up-to-date risk assessments and evaluation tasks on the work floor, ultimately contributing to occupational safety.

https://academic.oup.com/annweh/advance-article/doi/10.1093/annweh/wxaf081/8341514#541253998

Technologies used: LLM, NLP, NER, self-training AI models.

Deciding which drug target to push forward in the discovery of innovative therapies requires comprehensive analyses of large amounts of biomedical data, that are scattered and unstructured. The TargetTri platform tackles this issue with expert knowledge and data-driven approaches, allowing you to spend your time on strategic decision making rather than tedious data collection. Powered by custom-tuned large language models, expert knowledge and mined data, TargetTri uncovers crucial insights for assessing health outcomes and safety risks of food and drug interventions.

https://targettri.com/

Key technologies: Natural Language Processing.

This software tools helps industry to replace toxic chemicals by combining information on physicochemical properties needed for performance with toxicity information. AI-methods are used to fill data gaps.

https://www.tno.nl/en/safe-sustainable-by-design/

Technologies used: Privacy-by-design data analysis and modeling, generative AI, platform-as-a-service.

Phaeton is an ensemble of software packages developed with ambition to perform data analysis and modeling (DAM) on access restricted data. It is primarily developed to support DAM activities that need sensitive data at the time of a possible pandemic. It uses a combination of federated-learning like architecture and synthetic data to preserve privacy and generative AI for code safety and model generation.

https://phaeton.codeberg.page/