Turning Ideas Into Real-World Solutions
I am passionate about transforming scientific research and innovative ideas into practical technologies.
My projects span Artificial Intelligence, Cybersecurity, Machine Learning, Natural Language Processing, and Digital Entrepreneurship.
From developing AI-powered cybersecurity systems to building consumer-facing digital platforms, my goal is to create intelligent solutions that solve meaningful real-world problems.
Below is a selection of my entrepreneurial projects, research prototypes, and open-source contributions.
Entrepreneurship & Digital Products
Škrtica.ba — Smart Price Comparison Platform
Founder | Digital Entrepreneurship | 2026–Present
Škrtica.ba is a popular price comparison platform designed to make online shopping in Bosnia and Herzegovina more transparent, efficient, and affordable.
The platform collects and organizes product offers from hundreds of online stores, enabling consumers to compare prices, discover better deals, and make informed purchasing decisions.
Key capabilities:
- Automated collection and processing of product information from online retailers.
- Product matching and price comparison across different stores.
- Intelligent search and product discovery.
- Price tracking and shopping assistance features.
- Scalable data processing and integration of multiple e-commerce sources.
Škrtica combines software engineering, automated data collection, and intelligent information processing to help consumers save time and money.
My Role: Founder and Product Development
Website: www.skrtica.ba
Artificial Intelligence & Cybersecurity
TTPFShot — Few-Shot Cyber Threat Classification
AI Research | NLP | MITRE ATT&CK | Open Source
TTPFShot is a research framework for automatically identifying adversarial Tactics, Techniques, and Procedures (TTPs) from unstructured Cyber Threat Intelligence reports.
The framework combines Large Language Models with retrieval-based few-shot learning, using semantically similar examples to support classification with minimal labeled training data.
Key capabilities:
- Automated classification of MITRE ATT&CK techniques.
- Semantic retrieval of relevant few-shot examples.
- LLM-based classification of cybersecurity text.
- Evaluation of classification performance across CTI datasets.
TTPFShot demonstrates how retrieval-based learning can reduce dependence on extensive labeled datasets in cybersecurity applications.
My Role: Research and Development, in collaboration with AIT and TU Wien researchers.
Resources:
GitHub Repository | Scientific Publication
Advanced RAG for Cyber Threat Intelligence
LLMs | GraphRAG | Knowledge Graphs | Agentic AI
A research initiative investigating how advanced Retrieval-Augmented Generation architectures can improve question answering and reasoning over complex cybersecurity information.
The work systematically evaluates different approaches to retrieving and combining evidence from unstructured threat reports and structured knowledge graphs.
Research components:
- Traditional vector-based Retrieval-Augmented Generation.
- Knowledge graph-based retrieval and structured querying.
- Hybrid retrieval combining graph and textual evidence.
- Agentic retrieval strategies for complex analytical questions.
- Systematic evaluation of answer quality, factual grounding, and retrieval performance.
The objective is to support more reliable, explainable, and effective AI-assisted cyber threat intelligence analysis.
My Role: Research and Development, in collaboration with AIT and TU Wien researchers.
Resources:
APT Attribution — AI-Based Threat Actor Classification
Machine Learning | Cyber Threat Intelligence | Threat Actor Attribution
A research project exploring the application of machine learning and transformer-based language models to Advanced Persistent Threat (APT) attribution.
The goal is to analyze cybersecurity reports and identify patterns associated with known threat actor groups, helping analysts investigate potential connections between observed activities and adversarial behavior.
Key research areas:
- Classification of threat actor groups from cybersecurity text.
- Transformer-based language representations and semantic embeddings.
- Analysis of adversarial behaviors and MITRE ATT&CK techniques.
- Comparison of machine learning and language model approaches.
- Evaluation of model reliability and attribution uncertainty.
The resulting systems are designed to support analyst investigations rather than replace expert judgment or provide definitive attribution.
My Role: AI Research and Model Development
Resources:
AttackGroup-MPNET – Hugging Face
Open-Source AI Models
AITSecNER — Named Entity Recognition for Cybersecurity
Natural Language Processing | Transformers | Information Extraction
AITSecNER is a fine-tuned Named Entity Recognition model designed to identify cybersecurity-specific entities in unstructured text.
Built using the GLiNER architecture, the model supports the extraction of information such as malware names, threat actor groups, organizations, tools, and attack techniques.
Applications:
- Cyber Threat Intelligence processing.
- Automated entity extraction from security reports.
- Information enrichment and knowledge graph construction.
- Supporting downstream cybersecurity analytics.
My Role: Model Fine-Tuning and Development
Resources:
Explore More of My Work
My professional interests extend beyond the projects listed here. I continuously explore new applications of AI, machine learning, and intelligent software systems.
You can explore additional research, models, and implementations through my public profiles.
Let’s Build Something Meaningful
I am always interested in connecting with researchers, entrepreneurs, and organizations working on innovative applications of artificial intelligence and emerging technologies.
Whether it involves scientific collaboration, AI-powered products, or new technology ventures, I believe the most interesting opportunities emerge when research, engineering, and entrepreneurship come together.
