By Eni Muça — Emiliano Mankolli is one of those professionals who has brought a practical meaning to artificial intelligence in Albania. With a background in mathematical engineering and computer science, and a career that spans from academia to international startups, he now leads AI Hub Albania, the most active AI community in the country.
His newest project, “AI Deputy,” aims to make Albanian legislation more understandable and accessible by combining cutting-edge technologies such as machine learning, RAG, and agentic AI. In this interview, Emiliano shares his journey, what led him to this idea, and his vision of how AI can transform the legal field in Albania.
Can you tell us a bit about your background—what you studied and where your career has focused so far?
I studied Mathematical Engineering and Computer Science at the University of Tirana, Faculty of Natural Sciences. I completed my Bachelor’s and Master’s degrees in 2012 and 2014, respectively.
From the beginning, I was drawn to using mathematics to understand data behavior and build systems that could learn from it. At that time, the tech market in Albania was still developing, so I started my career in academia while constantly looking for ways to apply my knowledge in real-world contexts.
My first professional experience was as a lecturer at the Department of Computer Science at Aleksandër Moisiu University in Durrës. Although I enjoyed working in academia, I was eager to apply data analysis in the industry. My first experience outside academia was at a company focused on affiliate marketing, where we built an intelligent algorithm that automatically optimized ad campaigns.
In 2019, I joined Ritech International AG as Head of Data Science and AI—a role I held for over six years. The team I led worked on two highly challenging products.
The first project was for a recruitment startup, where we applied Natural Language Processing (NLP) algorithms to find the most suitable candidates for job positions. It was before the era of ChatGPT or large language models (LLMs), so the models required a lot of customization to fit real-world use cases. This experience sparked my interest in research, so in the same year, I began my PhD at the Academy of Sciences in Sofia. Four years later, I defended my thesis titled “Optimization Methods for Machine Learning Applications” and earned the title of PhD.
The second project involved computer vision and machine learning for one of the world’s largest transportation companies. The system enabled real-time monitoring and optimization of container unloading processes and was successfully implemented in several U.S. states.
In 2021, during my PhD, I received an offer from the Swedish startup Proxify, which connects software developers with clients. They implemented many of the algorithms I had developed, leading to a major improvement in candidate evaluation quality and recruitment speed. Within three years, the number of processed applications tripled, and the time from application to hiring decreased significantly. The company was later recognized as one of the fastest-growing startups in Sweden.
How did your interest in combining law and technology emerge?
In our field, it’s often hard to say which comes first—the problem or the solution. Sometimes you discover a new technology and wonder, “Where can I apply this?” Other times, you face an industry challenge and look for technological solutions to address it.
For example, while analyzing legal documents, we realized how time-consuming it is to interpret one article in the context of other laws. That’s when I clearly saw the potential role of AI in this domain.
So, the interest in merging law and technology came from two directions: our deep expertise in AI and machine learning models on one hand, and on the other, the huge untapped potential in Albania’s legal sector, where AI has barely been applied so far.
Which past professional experiences helped you the most in developing this project?
The field of data science and AI is like a puzzle—each new experience adds a piece to the bigger picture. The more you grow, the more you understand how these pieces connect.
Every project I’ve worked on has taught me how to build systems that understand context, not just words—an essential skill for any AI-based legal system. My experience with NLP models and Large Language Model development has been fundamental in creating the custom retrieval and reasoning algorithms we use in this project.
Have you previously worked on projects related to Albanian legislation or digital legal systems?
Not directly with Albanian legislation. The closest experience was developing a transformer model for embedding Albanian-language texts.
In English-language legal data, I worked on a project focused on contracts, where we built an unsupervised RAG graph to retrieve information across interconnected documents. In practice, this is a challenging problem because it requires multi-hop reasoning—the ability of an AI agent to understand links between multiple documents to provide a complete answer.
Tell us about your latest project. How did the idea for building a system combining RAG and Agentic AI for Albanian legislation come about?
I’m the co-founder of AI Hub Albania, the country’s largest AI community—both in terms of membership and collaborations with universities and companies. Our internal team includes 11 members, some of the top AI and data science experts working on international projects.
One of our core initiatives is called “Community Creates”, where the team develops AI-based products and makes them available to the community, public institutions, or companies. Through this program, we identified several high-potential areas for innovative AI solutions, and one of the most promising is “AI Deputy.”
This system is the first of its kind in the region and was introduced for the first time at a Parliamentary Research Conference, where it generated significant interest.
“AI Deputy” integrates several advanced AI and RAG technologies. Essentially, it’s a multi-agent system, where each agent specializes in a specific area and collaborates with others to provide accurate, well-reasoned legal answers.
Every user query creates a “virtual legal committee” that consults internally to generate the best possible response. The intelligent orchestrator decides which agents need to collaborate, allowing the system to analyze a legal issue from multiple perspectives.
For example, if a user asks about a specific Albanian law and wants to know how it differs from the corresponding EU regulation, the orchestrator brings together agents for both jurisdictions to generate a detailed comparison. This is crucial for aligning Albanian legislation with EU law.
Beyond that, the system can also be used in court cases, analyzing documents and providing a “verdict” based solely on the law. Another key feature is transparency—every answer is grounded in verified legal sources and is traceable, so users can see exactly which articles and laws were used to build the reasoning.
What has been the biggest challenge in building this system?
The main challenge was teaching the system not only to find a law but to understand the relationships between laws. That’s the heart of any intelligent legal system.
The most valuable component we’ve built is a custom algorithm that creates a graph structure allowing multi-hop reasoning. It combines retrieval with the ability to follow the logical chain of legal references.
This enables the system to gather all the information needed to give precise answers while minimizing the risk of hallucinations. The same solution could also serve as a foundation for many other public-sector AI applications beyond the legal domain.
Other challenges are human and financial resources. The project is self-funded by AI Hub Albania, including the setup of a mini data center, but to scale it further, we’ll seek collaborations with legal experts and financial support.
How do you see the role of this product in Albania’s legal services market?
This system can provide assistance to lawmakers and parliamentary committees by helping them cite, interpret, compare, or draft laws faster and more accurately.
It can support lawyers in quickly accessing and interpreting laws and managing case documentation. One of the greatest impacts could be in courts, where limited human capacity often causes long delays.
An AI system could process a large number of routine cases, significantly reducing costs and analysis time. AI doesn’t replace lawyers—it empowers them, giving them a faster, more accurate, and more accessible tool.
In the long run, collaboration with public institutions will be crucial to maximizing the system’s impact.
How do you plan to launch it in practice, and who will be the first users?
We aim to have a public version ready by November. There will be several user categories:
- General users, who can search for and explore legal information
- Legal experts, who will use it as an assistant for analysis and case management
- Parliamentary committees, who will use it for comparisons, citations, and legislative analysis
In the first phase, we’ll focus on collecting as much feedback as possible for improvement. Early users will essentially take part in co-creation, helping shape the system for real-world use.
What is your long-term vision? Do you think the system can expand beyond Albania or into other legal areas?
The system we’re building can become a legal expert in any knowledge base it’s given—whether it’s the Albanian Constitution, EU legislation, or that of another country.
The Albanian case is perhaps the most challenging due to the limited availability of language resources for Albanian. This makes the project even more valuable, as it pushes the boundaries of what AI can achieve in low-resource languages.
If we manage to achieve high accuracy here, there will be no barrier to adapting it for other countries.
For now, our focus is on Albanian and EU legislation, but the potential is vast—extending to public administration, education, and even economics.
0 comments