Backend developer• August 2021 - present
Developing backend systems for the largest online retailer in the Netherlands.
Developer Team Lead• February 2018 - August 2021
Guiding a team in the development of SaaS products using java/react. Developing, reviewing, mentoring and generally making sure everyone is moving in the right direction and working towards the feature goals. Also pushing the boundaries of what is possible with technology and how it can help us deliver a superior experience to our users.
Volunteer Teacher• June 2019 - present
Full-stack web developer• February 2017 - February 2019
Full-stack web development using java/react. I started organizing regular knowledge sharing sessions and coaching as part of a training program. On my own initiative I developed several reusable UI and back-end components using and integrating emerging technologies. These components allowed my company to gain a competitive advantage and increase satisfaction for our customers.
Freelancer• May 2015 - January 2017
Develop in-browser data visualizations using d3,flot,Charts,Highcharts. I focus on dynamic and interactive charts that can be updated with new data on the fly (ajax). I also work on projects involving natural language processing mainly with nltk and CoreNLP.
Researcher• September 2015 - February 2017
As a scientific Researcher at the MPI in Dortmund I studied the spatial organization in protein interaction networks by combining tools from machine learning and complex dynamical systems.
Researcher• June 2013 - August 2015
Designed and evaluated novel control tools for power systems, mainly under the smart grid and vehicle-to-grid concepts. Examined game-theoretic ideas that can reduce energy requirements for wireless networks by promoting cooperation. Gained a lot of experience with modeling software.
Researcher and Software Developer• July 2014 - July 2015
Worked on natural language processing by analysing micro-blog texts from Twitter and Kajgana forum. Also researched the social networks in Macedonia and used machine learning for link prediction on twitter. I was involved and headed several interesting projects in the above areas and others.
Teaching and Research Assistant• October 2012 - January 2014
Used Clus and Mulan to do very high-space mutli-label classification aiming to develop an automatic tag recommender for StackOverflow. Also taught undergraduate courses in the field of advanced programming practices and data structures.
Tool for searching the US patent database
Using d3 I developed an interactive way to explore the US patent database as a demo request for a client on Upwork. The tool allowed users to filter and group by multiple criteria, and follow the bubble chart to find the information they seek.
Managing Java Dependencies for Mendix Modules
I write about how to overcome the main issue when developing modules: dependency management. Using Gradle I show how one can deal with transitive dependencies & duplicate jars.
Electric vehicles assisting the power system during transients
Using the V2G concept which allows plug-in electric vehicles to feed power back to the grid we improve stability when the system is subjected to large disturbances. Verified through simulations using real data.
Sentiment analysis on forum posts in Macedonian
Classify posts as having a positive, negative or neutral sentiment on a topic. Users maintain a positive attitude when discussing food and shopping, but become grumpy when the focus shifts to the economy and global affairs.
Interactive visualization of child obesity rates in Macedonia
Obesity rates in different regions are presented using a color-coded interactive map. Percentage of kids who are overweight/obese are also divided by gender, ethnicity and regional development; or year 2010|2013.
Biology inspired cooperation in wireless networks
Applied biological mechanism such as network reciprocity to encourage cooperative behaviour in wireless networks. The result was reduced energy consumption for the group as a whole but also for individual nodes.
An interactive dashboard with custom sized and placed widgets
Building on top of a previous dashboard solution I allowed for widgets of different sizes. Users can drag around and reorder their widgets so that their favorite widget are in the spotlight. Preferences are automatically saved.
Decentralized vehicle-to-grid for improved steady-state stability of power grids
Local frequency measurements are all the information required to control a fleet of plug-in electric vehicles. Using a simple control function we increase system stability and robustness with respect to small disturbances.
Traffic infrastructure: planned underground speedway analysis
Applied tools from network science to foresee the effects of building a new speedway in Skopje. Average travel times are reduced and some current congestion sites disperse, but new ones are formed.
Bridging Online and Offline Social Networks: Multiplex Analysis
We study student friendships in real life and on Facebook, which come in two flavours: weak and strong. Based on this different types of network ties we define and analyse a multiplex network with 4 layers.
Just-in time Macedonian thesaurus and dictionary: for Word, Writer and PDF
A web service that provides definitions and translations for Macedonian words, and integrates information from Wikipedia on important places, people and more. Ready as a plug-in for MS Word, soon for Writer and PDF.
Analysing the temporal aspects of twitter networks
We looked at how twitter networks evolve over time by monitoring network metrics such as average distance, clustering, degree distr. etc. The network undergoes constant changes to keep in pace with current trends.
County Scholar: Visualizing census data
As per request from Tim Kubal I developed a soltuions using Highcharts that allows users to explore data from the US census. Users can select from different indicators; use different chart types: bar,line and area, and filter the data by county, state or whole nation.
Building hierarchies from flat tag clouds: important navigation aids
We examine how agglomerative clustering can help organize information on content-rich websites. Using only tag-cloud data we group related tags and/or tag-clusters, independently of the actual content.
Optimal sensor placement for pollutant detection in water networks
Showed how algorithms for sensor placement can be sped up by considering only a fraction of pollution scenarios. Demonstrated that network topology and centrality measures greatly influence node observational value.
Collection of digital resources in the Macedonian language
Developed a web application that allows fast customized searches using wildcards, bigrams or trigrams and lexicological meaning. Search implemented in lucene supported by regex matches for finding example usage.
Control architecture for power systems with plug-in electric vehicles
Design based on three-tier SoC: core operators, vehicle aggregators, individual car owners. Dedicated ethernet network with MPLS for low lag and reliability (tiers 1-2), supported by wi-fi (tiers 2-3).
Predicting your followers based on your tweets with machine learning
In order to get the predictions right we integrate both network data (followers) and text data (tweets). Looked at activity patterns, hashtag usage, tweet content, retweet frequency and how they influence befriending.
Promoting cooperation in selfish mobile wireless nodes
Using simple strategies borrowed from game theory, like tit-for-tat we showed that cooperative behaviour can emerge even from selfish decisions. Cooperation leads to both global and local energy efficiency.
Mood swings in opinions mined from micro-blogs
Investigate how public opinion about a subject changes with time, based on twitter data analysis. Found strong correlation between national basketball team wins-loses and twitter discussion thread.
Real-time tag recommendation
Case study: Stack Overflow
We build a two-level classifier. First level uses PCT (Clus) to choose good tag-clusters. Second level uses binary relevance (MULAN) with SVMs to give actual tag predictions within the clusters. Run-time < 100ms.
A method that measures a power characteristic at the one or more storage batteries, and controls the one or more batteries to selectively act as a generator to supply power to the power grid or to act as a load to draw power from the power grid in response to a predetermined disturbance in the power characteristic being detected.
Using recent results in the dynamics of coupled oscillators and delay differential equations we show that electric vehicles can be used to improve steady-state stability of power systems even in the presence of small communication delays. We corroborate the theoretical results with simulations on real power systems using PSS/E.
New Journal of Physics• 2014•PDF
We show that PEVs can improve stability when the power grid is subjected to large disturbances. We propose and test a control strategy that regulates the power output of a fleet of PEVs based on the grid frequency. The results show that: (1) speed and voltage fluctuations resulting from large disturbances can be significantly reduced up to 5 times, and (2) critical clearing time can be extended by 20-40%.
European Physics Journal ST• 2014•PDF
History and recent advancements in power system stability research. The classical and the structure preserving model characteristics and applications. Transient stability analysis by time-domain and direct methods. Steady-state stability analysis with application in designing novel control systems under the smart grid concept.
WiOPT 12th International Symposium, IEEE• 2014•PDF
Explored how simple network protocols can lead to cooperative behaviour in selfish nodes. The cooperation mechanisms that were set in place guarantee overall energy efficiency, while also maintaining balanced individual energy consumption. We concluded that cooperation can arise without complex rules that require data processing and memorization and little overhead communication.
Informatics and Information Technology, 10th Conference on• 2013•PDF
We investigate how correlated tags can be grouped by considering only the label part (output space) of the data. As a case study we looked into the StackOverflow tag cloud. We demonstrate that algomerative clustering algorithms perform really well in this setting, grouping similar and/or related technologies together e.g. "xcode" with "ios".
Transactions on Networking, IEEE/ACM • under review•PDF
We develop a game-theoretic framework to investigate the effect of cooperation on the energy efficiency in wireless networks. Two examples of network architectures are addressed, (1) ad-hoc networks and (2) networks with central infrastructure. The results indicate that uncomplicated local rules can promote network behaviour which yields both global and individual energy efficiency in certain wireless networks.
IEEE Transactions on Circuits and Systems 2 •under review•PDF
We make the system more robust to small disturbances using a smart grid concept that regulates generator power output based on remote frequency measurements. The measurements are transmitted using a dedicated wirelane communication network, thus ensuring low latency. Also developed an algorithm for optimizing the network infrastructure constrained by a given budget.
M.Sc. Computer Science • November 2014GPA 9.85
B.Sc. Computer Science • July 2013GPA 9.77