Science that leaves the lab*

* and reaches people, patients and companies.
We run EU-funded projects with universities, hospitals and companies across Europe, and publish what we learn. The same methods end up in the pilots and products we build for clients.

0
EU-funded projects
0
scientific publications
0
open datasets
0
academic and clinical partners
EU-funded research

Three programmes, one aim: research people can use

We join European consortia as the partner that builds, pilots and trains. Each project pairs universities and public bodies with a team that ships working systems.

Horizon Europe

SMART Researchers

2025 to 2028 · Running

An 11-partner consortium helping early-career researchers build the skills and careers they need, through career hubs, training and links with industry in Croatia, Serbia, Greece and Latvia. As the industry partner, Sparky* runs training, events, mentoring and industry connections in Croatia.

Erasmus+ Alliances for Education and Enterprises

ESG4PMChange+

2024 to 2027 · Running, pilot phase

Fourteen partners bring environmental, social and governance thinking into project management education. Sparky* supports the curriculum and training, work-based learning, Living Lab projects, mentoring and dissemination.

I3 Interregional Innovation Investments

HealthChain

2023 to 2025 · Completed, moving to market

As Leading SME for the FallPredict challenge with KBC Rijeka, Sparky* built StabilityGuard: medical-grade wearable sensors, a monitoring dashboard and alerts that predict patient falls, processed on-premise in the hospital. The pilot took the system from TRL 6 to TRL 9.

From the projects

Workshops, pilots and partner meetings

Photos from the official project websites: smartresearchers.eu, esg4pmchange.com and healthchain-i3.eu.

NVIDIA Inception

Built on accelerated computing*

* Sparky* is a member of NVIDIA Inception, the NVIDIA programme for startups. We use NVIDIA accelerated computing across our robotics R&D, edge AI and cloud work, from training models to running them in real time.

Robotics

Our own AI for open-source robots such as Reachy Mini and Microduck: perception, speech and behaviours trained in simulation and tested on NVIDIA GPUs.

Edge AI

Models that run next to the data, on wearable sensors, cameras and drones, so analytics happen in real time where the signal is.

Cloud and real time

GPU-accelerated training and inference in the cloud for real-time video, streaming analytics and GenAI workloads.

Scientific publications

Featured peer-reviewed work

A selection from the 50+ papers our team has published, in text mining and big data, AI and the economy, medical imaging and computer vision. Sparky* authors are highlighted.

0
featured papers
0
citations of these papers on Google Scholar
2020
Text mining of Industry 4.0 job advertisements

M. Pejić-Bach, T. Bertoncel, M. Meško, Ž. Krstić
International Journal of Information Management 50, 416–431 · cited 383 times

2019
Text mining for big data analysis in financial sector: a literature review

M. Pejić Bach, Ž. Krstić, S. Seljan, L. Turulja
Sustainability 11 (5), 1277 · cited 349 times

2019
Visualization of big data text analytics in financial industry: a case study of topic extraction for Italian banks

Ž. Krstić, S. Seljan, J. Zoroja
ENTRENOVA Enterprise Research Innovation 5 (1), 35–43 · cited 27 times

2019
Big data text mining in the financial sector

M. Pejić Bach, Ž. Krstić, S. Seljan
Expert Systems in Finance, 80–96 · cited 23 times

2019
Big data for prediction: patent analysis, patenting big data for prediction analysis

M. Pejić-Bach, J. Pivar, Ž. Krstić
Big Data Governance and Perspectives in Knowledge Management, 218–240 · cited 18 times

2014
Big data and sentiment analysis using KNIME: online reviews vs. social media

A. Minanović, H. Gabelica, Ž. Krstić
37th International Convention MIPRO 2014 · cited 50 times

2024
Economic theory and artificial intelligence: a cross-model perspective on labour market dynamics

Ž. Krstić
Croatian Regional Development Journal 5 (2), 52–75 · cited 16 times

2023
Predictive maintenance in Industry 4.0 for the SMEs: a decision support system case study using open-source software

M. Pejić Bach, A. Topalović, Ž. Krstić, A. Ivec
Designs 7 (4), 98 · cited 65 times

2015
Grassroots campaign as a technique of indirect lobbying

I. Bilić, I. Vuković, Ž. Krstić
Communication Management Forum 2015

2025
Age estimation through sternal fusion and costal cartilage ossification using MSCT in a Croatian population: model development and application

J. Vickov, I. Jerković, I. Perić, N. Jerković, Ž. Bašić
International Journal of Legal Medicine

2022
Bipartite parietal bone: a case from Croatian population

Ž. Bašić, I. Jerković, T. Bareša, N. Jerković, et al.
Journal of Bioanthropology

2024
Comparing YOLOv5 and YOLOv8 performance in classification tasks (Usporedba performansi YOLOv5 i YOLOv8 algoritama u zadacima klasifikacije)

E. A. Drašković
Undergraduate thesis, University of Split, Faculty of Science (in Croatian)

Open science

Datasets anyone can reuse

Research data from our projects, cleaned, documented and published under CC BY 4.0 so other researchers can test, compare and build on it.

0 .190
ESG job postings

ESG LinkedIn Jobs Dataset

Job postings from 2.827 companies, August 2025 to January 2026, with full descriptions for NLP, skills and labour market research.

CC BY 4.0
JSON
CSV
Parquet
0 .736
sensor measurements

HealthChain Sensor Data

9-axis motion data at 52 Hz from three medical-grade wearable sensors, recorded during a 55-patient clinical study at KBC Rijeka. Anonymised and GDPR compliant.

CC BY 4.0
CSV
Parquet
52 Hz
Partners

Universities and clinics we work with

Students and young researchers

Research careers start here*

* Internships, student placements and research projects with the universities of Split, Zagreb and Edinburgh, students from across Europe through SEA-EU, and ESG and responsible AI work with 3PSplit.

Featured research project · University of Edinburgh

Do GenAI agents cite what they claim?

Second-year PhD students from the University of Edinburgh’s Centre for Doctoral Training in Designing Responsible NLP work with our team on the faithfulness of citations in advanced GenAI research agents. When an agent writes a detailed report from documents, database tables and the web, is every claim supported by the evidence it cites? The students audit each citation against the exact evidence the agent received.

The study runs 200 detailed reports over 100 job postings from our open ESG dataset, with probes for known failure modes, and keeps a sanitised trace of every run so each claim can be checked by hand.

Recruiter or employer
Mention or requirement
Numeric claims
Table-only facts
Duplicate postings
Regulation context from the web
0
reports audited claim by claim
0
ESG job postings as the evidence base
0
evidence types: documents, tables, web
Research partnerships

From hypothesis to evidence*

* Planning a proposal or a study that needs a team that collects the data, builds the models and pilots the result? We join consortia as the partner that turns research into working systems.

Horizon Europe
Erasmus+
I3 Interregional Innovation
National research calls