Summary
- Profile Type
- Technology offer
- POD Reference
- TOHR20260810010
- Term of Validity
- 13 August 2026 - 13 August 2027
- Company's Country
- Croatia
- Type of partnership
- Research and development cooperation agreement
- Targeted Countries
- All countries
General information
- Short Summary
- Croatian micro-enterprise specialising in applied artificial intelligence and machine learning offers know-how in language models, agent-based systems, computer vision, speech and traditional machine learning, covering the full path from feasibility assessment through prototyping to production-grade models. It seeks a research and development cooperation agreement with organisations preparing European project proposals that include artificial intelligence work and need a partner to deliver it.
- Full Description
-
Croatian micro-enterprise (sole proprietorship, one employee at the moment) providing artificial intelligence and machine learning engineering and consulting services, active since 2023. Its founder, Mislav Jurić, holds an MSc in Computer Science from the Faculty of Electrical Engineering and Computing, University of Zagreb, has more than five years of professional experience across industry and applied research, and has co-authored peer-reviewed publications. The company is registered in the EU Funding and Tenders Portal and its founder is registered in the European Commission's database of independent experts for the evaluation of proposals.
Many research and innovation projects now include an artificial intelligence component, but consortia frequently lack a partner able to carry a model from a research prototype to something that runs reliably in production. Prototypes stall at demonstration stage, models are trained without a validation regime that survives release, and effort is committed to targets the available hardware cannot meet. The know-how offered here addresses that gap.
The specific know-how on offer covers: optimising transformer models for production, including an encoder layer-pruning technique that reduced inference time of a BERT-based production model by approximately 50%; building natural language processing models end to end, from dataset design and labelling assisted by large language models (LLMs) through training to integration into a live product; fine-tuning open-weight models and running them, together with speech recognition and synthesis, on local, on-premise and edge hardware; hardening LLM-based applications against prompt injections and system prompt extraction, and adding content filtering and safety guardrails; evaluation, benchmarking and differential testing under strict validation standards before release; computer vision, including semantic segmentation, object detection, pose estimation and analysis of aerial and medical imagery; and early feasibility assessment and architecture definition, establishing whether artificial intelligence is the right tool at all before development commits to it.
This know-how has been applied in commercial software, video games, advertising analytics, precision agriculture, healthcare imaging and the automotive sector, and is transferable to any application field.
A research and development cooperation agreement is sought because the company works best as a technical partner inside a larger project rather than as a supplier of a finished product. In practice this means joining a consortium as a beneficiary that owns defined artificial intelligence tasks within one or more technical work packages: feasibility assessment and architecture definition in the early phase; dataset creation, annotation strategy and data curation; development, fine-tuning and optimisation of models; evaluation, benchmarking and validation, including safety and robustness testing; integration of prototypes into production-grade services and pipelines; and contributions to dissemination, training and exploitation activities. Involvement from the proposal-writing stage is preferred, so that the technical work packages can be shaped realistically, but the company can equally join a consortium that is already formed and has a specific gap to fill. - Advantages and Innovations
-
• Encoder layer pruning applied to a BERT-based production model cut inference time by roughly 50% without a separate model redesign, in both the production and the model training frameworks
• Prompt-protection work on a commercially released product raised time-to-extract the system prompt from under one minute to roughly 30 minutes for technical testers
• Full coverage of the project lifecycle in a single partner - feasibility assessment, prototyping, model development, evaluation and production integration - closing the usual gap between a research prototype and a deployable system
• Production discipline: experience maintaining long-lived production pipelines, retraining models under strict validation standards and running differential testing before release
• Experience running models locally, on-premise and on edge hardware, relevant to projects with data sovereignty or privacy requirements, and to identifying hardware limits before they become project risks
• Participation contributes to consortium balance: a micro-enterprise from a widening country, with low administrative overhead and fast decision-making
• Remote-first working, fluent in English and Croatian - Technical Specification or Expertise Sought
-
Areas of particular interest
The company is open to any project with an artificial intelligence or machine learning component, in any application sector and under any funding programme. Two areas are of particular research interest — generative artificial intelligence for 2D and 3D animation and creative production pipelines, and trustworthy and safe artificial intelligence, meaning evaluation, benchmarking, guardrails and red teaming — but these are stated as interests rather than as conditions, and enquiries outside them are equally welcome.
Selected track record
As a one-person company, the capacity offered is that of its founder, Mislav Jurić. The engagements below cover both client work delivered through the company (2023-present) and his earlier employment.
• Senior Machine Learning Engineer, Newfire Global Partners (2025-present): reduced inference time of a BERT-based production model by approximately 50% through encoder layer pruning; built a production natural language processing model end to end, from dataset design through training to integration into a live product; retrained models under strict validation standards with differential testing before release
• Machine Learning Engineer, GoodAI - AI People (2023-2025): worked on a commercially released computer game driven by LLMs; fine-tuned local 7B models, integrated local models and local and cloud speech recognition and synthesis, engineered prompt-protection mechanisms, and built content-filtering and safety mechanisms
• AI Lead, computer vision advertising analytics project (2025, confidential client): defined the technical architecture with the client's executive and technical leadership, identified that the selected edge hardware could not run the intended pipeline and prompted a hardware change before development committed to an unworkable target, built the project's first complete end-to-end pipeline, and mentored a junior engineer
• Bug triage and root cause analysis agent (2026, confidential client): delivered a proof-of-concept agent-based pipeline chaining a no-code agent into a prompt-driven coding agent
• Machine Learning Engineer, TIS Group (2020-2023): led a drone-imagery pipeline from design to cloud deployment for precision viticulture, segmenting vineyard rows and gaps, estimating vine-gap counts and assessing overall vineyard health, presented jointly with the agricultural partner at the AI2FUTURE 2022 conference; earlier, reduced pose-estimation model training time from weeks to days through GPU and multi-GPU training and improved accuracy through transfer learning
• Dissemination and training: lectures at AI Center Lipik, and an invited public talk on preserving cultural heritage in the age of artificial intelligence
Technology stack
• Domains: GenAI, LLMs, NLP, Computer Vision, Traditional ML, Agentic AI, Data Science, RAG, ASR, TTS
• Programming languages: Python, C++, SQL
• Libraries and frameworks: PyTorch, Hugging Face, scikit-learn, FastAPI, NumPy, OpenCV, FlairNLP, spaCy, pandas, LlamaIndex, Streamlit, Keras, SciPy, PyInstaller, TensorFlow
• Databases: CosmosDB, Chroma, PostgreSQL
• Tools and cloud: Azure, Git, Jupyter, Linux, CI/CD pipelines, Jenkins, W&B, GCP, Vertex AI (incl. ADK), Docker, GNU Make, Terraform, Prodigy - Stage of Development
- Already on the market
- Sustainable Development Goals
- Goal 9: Industry, Innovation and Infrastructure
- IPR status
- No IPR applied
Partner Sought
- Expected Role of a Partner
-
Sought: universities, research and technology organisations, companies and public bodies with experience of EU-funded projects, that are preparing or have already begun preparing a proposal which includes artificial intelligence or machine learning work and need technical capacity to deliver it. Relevant programmes include Horizon Europe (particularly Cluster 4 - Digital, Industry and Space), Digital Europe, Eurostars-3 and Interreg, but the company is open to any EU-funded or nationally co-funded research and innovation programme.
The expected role of the partner is to act as coordinator, or as a consortium partner with influence over consortium composition, and to include the company as a beneficiary responsible for defined artificial intelligence tasks within one or more technical work packages. The company does not seek to coordinate. Contact from the proposal-writing stage is preferred, so that the technical work packages can be shaped realistically, but a consortium that is already formed and has a specific technical gap to fill is equally welcome. Participation as a linked third party or as a subcontractor is possible where the consortium structure requires it. - Type and Size of Partner
- SME 50 - 249
- SME <=10
- University
- R&D Institution
- Other
- Big company
- SME 11-49
- Type of partnership
- Research and development cooperation agreement
Dissemination
- Technology keywords
- 01003003 - Artificial Intelligence (AI)
- 01003012 - Imaging, Image Processing, Pattern Recognition
- 01003017 - Speech Processing/Technology
- 01005004 - Human Language Technologies
- 01003001 - Advanced Systems Architecture
- Market keywords
- 02007016 - Artificial intelligence related software
- 02007020 - Artificial intelligence programming aids
- 02007021 - Other Artificial intelligence related
- Targeted countries
- All countries
Files
https://www.mislavjuric.com/