Anna Tshngryan

An automated machine learning tool as a helper to data scientists and machine learning specialists. I was the core person developing the project from the initial until the last stages.

The tool accepts the dataset and outputs consistency issues if present, data cleaning tips, provides comprehensive dynamic graphs, gives the ability to clean the data with only one click, and later trains the machine learning model.

The user chooses either default cleaning ways the tool offers or gives custom values and decisions. Ability to train the model is provided. The tool gives some tips and precalculated values for predictive model parameters; however, the user is free to choose any model and parameter value.

This ML tool became a great way of optimizing the company's AI teamwork, as it speeded up our work by more than 20%.

A machine learning pipeline that gives insight to the Bank into whether the new customer will be able to pay the loan back or not.

I was responsible for this project from the first stage of its development. I conducted data analysis and provided reports to the Bank. I trained multiple machine learning models, compared them, and checked for possible biases.

I developed the whole pipeline. It takes raw data concerning the user as input, encodes it, and gives it to the predictive model. The model outputs the probability of the person paying back the loan.

Built a technology that finds logos in images and gives corresponding locations. The software also works with the videos. It provides the timestamps of the emblem of interest in a video.

I developed the project from the first to the last stage, including data gathering and scraping, cleaning, model development, and testing.

The project aimed to create a robust and scalable mentor recommendation system that matches mentees with suitable mentors based on their profiles. This system combined rule-based logic with autoencoder techniques to enhance the accuracy and relevance of recommendations.

KEY FEATURES
• Leveraged profile embeddings that encode user information into a vector space.
• Incorporated multiple factors into a rule-based system. Profile embeddings were added as a component to enrich the scoring system.
• Processed and cleaned large datasets, removing outliers to ensure training data quality.
• Conducted a thorough analysis comparing two phases of the system's development. Established benchmarks, tracked precision at top-N recommendations, and identified areas for optimization.
• Partnered with product and development teams to define success metrics and worked with designers to ensure a user-friendly interface.

OUTCOME
Delivered a scalable, data-driven recommendation system that aligns with business goals and enhances user satisfaction through tailored mentorship matches.