Marko Arezina

Experience with

Multimodal Amazon Alexa, Google Assistant, and Samsung Bixby experiences.

For brands, I worked on creating new revenue streams with In-Skill Purchasing which yields higher conversions or augmenting an existing app by creating a voice interface powered by the existing backend.

For news sites, I worked on leveraging voice assistants to create a new distribution channel for content, increasing personalization, and sponsor impressions. I’ve worked extensively with Python, Neo4j, and Google Cloud to develop the core explainable content recommendations technology that powers Vecgraph’s voice experiences for news sites.

I've worked extensively with Node.js voice frameworks such as JOVO and tools like SSML.

Machine learning models and natural language processing

Experience developing models for various problems, including computer vision, natural language processing, and reinforcement learning.

A few techniques I have used in the past: XGboosting, convolutional NNs, LSTMs, word embeddings, Q-learning. Both supervised and unsupervised learning.

Experience developing machine learning models with Tensorflow and Keras.

I worked on a Natural Language Processing pipeline for extracting entities, enriching entities with metadata using IBM watson, and applying machine learning models to create document embeddings.

Web apps

I've got experience working with React, Gatsby, and Webpack to develop numerous web apps.

Designed a visual content discovery tool to allow readers to further explore a topic and discover more relevant content on a blog or news site.

Developed "Slide to explore" experience which allows users to visualize topic clusters, as well as see how the content recommendations are coming from the underlying graph.

Content recommendation algorithms

I am a Neo4j Certified Professional

Experienced creating complex Cypher Queries and working with Knowledge Graphs.

Worked on a system to provide contextual and explainable content recommendations for blogs and news sites.

Used a Graph database (Neo4j) to develop recommendation algorithms as well as apply various clustering and pathfinding algorithms for finding topics and curating content.

Projects

Tech News Explorer

Every week there seems to be an overwhelming amount of things happening in the tech world. New startups, products, funding rounds, acquisitions, and scientific breakthroughs. I wanted to create a way to visualize and explore the connections between stories that shaped this week in tech.
Given my love for graphs and machine learning I thought it would be fascinating to visualize the stories in a graph. Every week, the app analyses the connections between the latest stories using a combination of machine learning and graph algorithms.

React
HTML5
Neo4j
Python (NLP)
Flask

Health News Action

Health News is a Google Assistant Action designed for exploring health stories. This voice app showcases some features that can be used to create voice experiences for news and media sites. It was build with Node.js (Jovo) framework.
The the articles are sourced from the Harvard Health Blog, and all of the text content is analyzed automatically. Here is my post outlining some of the ways a voice app can be used to help blogs, which served as inspiration for this voice app.

Node.js (Jovo)
Dialogflow
Neo4j
Python (NLP)
SendGrid

A fun project. It is a lyrics generator that can generate lyrics in the style of over 40 artists. The model is trained on lyrics from Lyrics Beast. You can also specify the target number of syllables.

React
Python
Tensorflow
Flask

Exploring 16000 Frontiers in Neuroscience Articles

The recent interest in Neuralink is the inspiration behind this analysis. I wanted to explore what are the some of the trends in brain computer interface research. Frontiers Media is an academic publisher of peer-reviewed open access scientific journals. This project is an exploratory data analysis of 16,000 Frontiers neuroscience journals.

Python
Pandas
Matplotlib
Jupyter Notebook

Bio-monitor Activity Prediction

Created a model to predict if a subject is walking, run/jog, biking, working on computer, watching TV, sleeping, or eating.
Worked with a challenging dataset with biometric features across multiple csv files.
Activity meta data annotated using natural language with many sparse categories.
Stitched together the biometric data and labeled with correct activity based on time interval.

Python
Pandas
XGboost

About Me

I am a fullstack developer based in Ottawa, ON. I am passionate about keeping up with the latest developments in voice assistants and machine learning. After seeing tremendous growth in voice assistant in the last few years, I'm excited about the future of voice technology startups and the potential for entire platforms to emerge, built on top of voice assistants. In my spare time I enjoy creating projects to explore interesting new technologies.

Resume

Get In Touch

I help brands build their Google Assistant and Alexa experiences. If you're thinking of taking advantage of this ever growing opportunity but are not sure where to start, let's get in touch so we can figure out how voice can work for you.

Email