Didi Milikina

Amsterdam, the Netherlands

My job as a Consultant at DeepLearning.ai was to review the videos and test the notebooks from the specialisations before publishing them in Coursera. Additionally, I discussed necessary edits regarding the content with the corresponding teams. The mistakes and suggestions were added to Trello boards which were reviewed by the editors. I had worked on the NLP track, the TensorFlow track and partially on the GAN's track.
My goal was to improve the notebooks, which consisted of expressing the requirements in an understandable way for the end-user and adding flowcharts or hints whenever needed.

Sofia, Bulgaria

Being an ambassador of Coding Girls enabled me to organise and facilitate tech-related events, be a speaker, and take part in important IT events. The events which I organised were for different age groups. Most of them were for women, but I also organised workshops for kids. Supporting more girls to follow their dreams was one of my main objectives. Companies that I worked with were Uber, Leanplum and MindHub, and I attended many events and conferences as an ambassador to represent the organisation and its values.

Sofia, Bulgaria

At DevriX I made plugin for WordPress, used WordPress admin panel for adding new info for sites. Updated some of the components which were in different projects and implemented new ones. Used Sass (Sassy CSS) and JS. During the intern, I accumulate basic knowledge of PHP which we were used for back-end for our plugin.

Sofia, Bulgaria

At JBoxers I improved my personal JS and React skills with different courses from egghead.io and Udemy. I did small projects to consolidate what I learned and also applied it in the company.

This specialization helped me build a solid foundation for my understanding of Machine and Deep Learning. It helped me to learn the best practices and gave me many insights from recognized and influential people in the AI field. It also prepared me for getting more into the depths of different applications and approaches towards Machine Learning.

Certificates

By completing this career track, I gained more knowledge and experience with Python. There were many coding assessments which covered abundant aspects of Data manipulation, the Python library - pandas(manipulating and merging Data Frames), relational databases in SQL, Intro to Shell and Conda essentials.

This skill track helped me familiarise myself with one of the most popular BI tools, namely - Tableau. I learned how to organize and analyze data, create presentation-ready visualizations, build insightful dashboards, and apply analytics to worksheets. Moreover, I used data connectors to combine and prepare datasets and combined multiple data tables with various relationships, joins and unions. Furthermore, I learned how to manage different data properties, like renaming data fields, assigning aliases, changing data types, etc.

This career path successfully enriched my insight into Deep Learning. The track contained more intermediate courses on SQL and Python libraries such as Matplotlib and Seaborn, where I covered a wide variety of displaying the data(making subplots, overlaying plots, as well as strip, swarm and violin plots). Furthermore, there were two courses for Statistical Thinking in Python, where I cleared the previously acquired data to make an accurate conclusion for the tendencies.

Finishing the Data Scientist career path, I successfully upgraded my skills from the last two tracks(Data Analyst and Python Programmer) thanks to the project and the case study I solved.

Completing the Data Engineer track, I deepened my expertise in building datasets from imported files with different file formats and Bash scripting. Moreover, I learned how to process data in Shell and did command-line automation.
I integrated Amazon Web Services (AWS) into data workflow, uploading data to Boto Amazon Simple Storage Service (S3), creating buckets, and subscribing people to SNS to receive critical notifications via SMS. Boto is an AWS software development kit for Python that enables creating, configuring and managing AWS services. Amazon Simple Notification Service helps manage services that deliver messages from producers to consumers.
Another topic was Big data analysis with Apache Spark. Spark helps you interface with RDDs. PySparkSQL is a library to apply SQL analysis, while the MLlib machine learning library is a wrapper over PySpark.
Another discovered area was error handling, transactions in SQL Server, and building and optimising triggers in SQL Server.

This track gave me significant knowledge about Keras, a Python library capable of running on top of TensorFlow.
There were exercises about building Sequential models with Dense layers(different numbers of neurons and layers). Also, I used Keras for image processing and classifying clothing types. XGBoost is a fast and scalable gradient-boosting library.
I learned how to measure AUC(Area under the curve) and accuracy, import hyperparameters with a Pipeline and many other valuable usages.
Another important Machine Learning tool included in the path was Apache Spark, integrated with Python.

January 2023

Scientific paper

This project aims to investigate the influence of location and price on restaurant ratings, addressing the lack of research in this area. By utilizing advanced statistical techniques, the project seeks to fill this gap and provide valuable insights for the restaurant industry.

Methodology:

• Cluster Analysis: Employing cluster analysis, the study categorizes restaurants based on their proximity to city centers. This analysis uncovers patterns and clusters within the data, revealing how city center proximity plays a role in influencing restaurant ratings.
• Multiple Linear Regression: The project utilizes multiple linear regression to assess the main and moderating effects of location and price on restaurant ratings. This technique helps establish relationships between these factors and their impact on ratings.

• Cluster Insights: The analysis uncovers four distinct clusters of restaurants based on their city center proximity. This categorization provides valuable insights into how geographical location influences restaurant ratings.
• Significant Impact: The research demonstrates that the identified clusters significantly affect restaurant ratings. The project's findings shed light on the previously unexplored relationship between location and ratings.
• Price Moderation: The study reveals that price has a moderating effect on the relationship between location and restaurant ratings. Specifically, the effect of location on ratings varies depending on the pricing strategy adopted by restaurants.

April 2022

Scientific paper

This project investigates strategies for managing imbalanced image datasets in Convolutional Neural Networks (CNNs) by comparing two techniques: oversampling and class weighting. The focus is on a dataset containing images of whales and dolphins, which exhibits a significant class imbalance.

Objectives:

• Compare the effectiveness of oversampling and class weighting methods for imbalanced data in CNNs.
• Evaluate the performance of three residual convolutional models with variations in end layers, depth, and width.

• Employ three residual convolutional models incorporating modern architectural features such as residual connections, global average pooling, and swish activation.
• Implement oversampling to duplicate instances from minority classes and utilize class weighting to assign appropriate importance to each class during training.

• The experiments demonstrate that oversampling leads to higher accuracy in the classification task.
• However, oversampling also introduces a higher false-positive rate for classes with limited instances.
• The choice between oversampling and class weighting depends on striking a balance between accuracy and true-positive rate, particularly in real-world production scenarios.

February 2018 - July 2018

A platform inspired by the BookCrossing initiative and the trends of a shared economy.
The main idea is to give people the possibility to connect with others that have a particular book in a given region and exchange one book for another without actually putting any monetary value in the exchanged books. This way, small communities can be created within cities/regions/counties or even within a company/school or university.
The biggest goal was to make the most major open library that enables books to travel worldwide.
For the front-end we used React.js with the material design MDL Lite library for styling the interface.
For the back-end we used ASP.NET Core with EF Core and SQL Server.

Demo website - The demo website currently has a lot of placeholder info and media until the actual "content" is poured in.

April 2018 - October 2018

We created the website platform to show our development skills. The main idea was to find diverse and exciting projects. We were motivated and open to new challenges. That was the reason for creating the company which offered such types of services. It provided users to specify and customise their needs for a web platform.

Demo website - The demo website currently has a lot of placeholder info and media until the actual "content" is poured in.

Arduino based self watering planter

November 2018 - February 2019

We were motivated to take care of our plants and wanted to give them precisely calculated water and sunlight, without which they could not survive. Some of the species are very demanding and difficult to grow. So we decided to make our custom garden project to monitor them.
We used Arduino to receive and display the values with different components.
For the database, we used Apache Cassandra; for displaying the charts, we decided to use React-Vis, and Python for data ingestion from Arduino through USB (Future versions through LoRaWAN), mapping the input and updating Apache Cassandra and GoLang - REST/GraphQL (not yet decided and implemented) from the Apache Cassandra.

GitHub project