Vahe is available for hire
Hire VaheVahe Zakaryan
Verified Expert in Engineering
Data Scientist and Developer
Yerevan, Armenia
Toptal member since April 8, 2022
Bio
Vahe is a seasoned data scientist and machine learning scientist with 8+ years of IT experience and a software engineering background. He has been primarily involved in building and developing an in-house search engine, recommendation systems, and demand and supply predictions. His most recent projects are related to generative AI, such as Text2Image (Stable Diffusion) and LLMs (ChatGPT). Vahe is an expert in machine learning and deep learning and is keen on leading small teams.
Portfolio
SumerSports, LLC - Main
Python, Data Science, APIs, Agentic AI, OpenAI, Vector Databases, LangChain...
ServiceTitan
Machine Learning, Deep Learning, Recommendation Systems...
X49 FZ LLC
Python, Machine Learning, Recommendation Systems, Social Networks, Search...
Experience
- Python - 7 years
- Deep Learning - 6 years
- Machine Learning - 6 years
- Data Analysis - 5 years
- Data Science - 5 years
- PySpark - 4 years
- Recommendation Systems - 3 years
- Search - 3 years
Preferred Environment
Slack, PyCharm, Databricks, Hadoop, Jupyter Notebook, Terminal, Amazon Web Services (AWS)
The most amazing...
...solution I've developed is an in-house search engine that includes different types of machine and deep learning sub-projects and an R&D testing environment.
Work Experience
Python Data Scientist (via Toptal)
2023 - PRESENT
SumerSports, LLC - Main
- Specialized in the development and optimization of data and model validation pipelines, ensuring data integrity and maximizing machine learning model performance.
- Implemented strategic approaches for rigorous model validation, including iterative testing and continuous improvement to achieve optimal results.
- Collaborated seamlessly with cross-functional teams to unify data and model validation processes, contributing to the overall success of data science initiatives.
Technologies: Python, Data Science, APIs, Agentic AI, OpenAI, Vector Databases, LangChain, Cloud Services, REST APIs, Testing, Reliability, Design, Speech-to-Text (STT), AI Programming, AI Modeling, XGBoost, Statistical Methods
Senior Data Scientist
2023 - PRESENT
ServiceTitan
- Worked on the chatbot project based on generative AI and large language models (LLM), which aimed to understand a customer's issue, collect all necessary information, and book an appointment.
- Developed a recommendation system based on the LLMs.
- Engaged in the research and development of personalized LLMs based on transfer learning and model tuning.
Technologies: Machine Learning, Deep Learning, Recommendation Systems, Natural Language Processing (NLP), Language Models, Generative Pre-trained Transformer 3 (GPT-3), Generative Pre-trained Transformers (GPT), Model Context Protocol (MCP), Large Language Models (LLMs), Agentic RAG Systems, Prompt Engineering, Retrieval-augmented Generation (RAG), APIs, Agentic AI, OpenAI, Vector Databases, Workflow Automation, LangChain, Cloud Services, REST APIs, RAG Pipelines, Testing, Reliability, Design, Snowflake, Speech-to-Text (STT), AI Programming, AI Modeling, XGBoost, RAG Systems, Generative Artificial Intelligence (GenAI), Statistical Methods
Machine Learning Team Lead
2025 - 2025
X49 FZ LLC
- Developed social network recommendation system architectures and system designs.
- Managed a team of seven people, fostering the development of teamwork, ownership, and leadership skills.
- Created automation systems for the deployment and monitoring of recommender systems (RecSys) engines.
Technologies: Python, Machine Learning, Recommendation Systems, Social Networks, Search, Artificial Intelligence (AI), Social Media, LangChain, Cloud Services, REST APIs, RAG Pipelines, Testing, Reliability, Design, Speech-to-Text (STT), AI Programming, AI Modeling, XGBoost, RAG Systems, Generative Artificial Intelligence (GenAI), Anthropic
AI Consultant via Toptal
2023 - 2023
Asym Labs Limited
- Contributed to strategic AI implementation. Orchestrated the strategic integration of AI solutions in a startup environment, aligning technologies like large language models (LLMs) with business objectives.
- Customized LLM applications. Developed tailored LLM applications, including ChatGPT, to address specific client needs, ensuring project success and exceeding client expectations.
- Contributed to knowledge transfer and team empowerment. Led knowledge transfer sessions to upskill the team on AI advancements, fostering a culture of continuous learning and enhancing internal expertise for sustained innovation.
Technologies: Artificial Intelligence (AI), Natural Language Processing (NLP), OpenAI GPT-3 API, Llama 2, Claude, Large Language Models (LLMs), RAG Pipelines, Testing, Reliability, Design, AI Modeling, XGBoost, Generative Artificial Intelligence (GenAI)
Senior Machine Learning Engineer
2023 - 2023
AIBY
- Worked on tuning stable diffusion models to make generations more realistic and without artifacts.
- Researched the fields of generative AI and image generation.
- Developed a Text2Image model that can generate aesthetic images with high resolution.
Technologies: Machine Learning, Deep Learning, Stable Diffusion, Computer Vision, Image Generation, APIs, Cloud Services, Testing, XGBoost
Deep Learning and Python Developer
2022 - 2022
KYROS Technologies LLC
- Developed a deep learning solution to predict the demand for the used bonus points for clients and predict their behavior in 30-50 months. The model decreased the costs and increased the revenue of the clients.
- Worked closely with the CEO to analyze the data and the potential issues.
- Discovered the problems in the previous solution and optimized them.
Technologies: Deep Learning, PyTorch, Python, Apache Spark, Spark, ETL, Machine Learning, Finance, Data Engineering, Distributed Systems, GitHub, APIs, Cloud Services, Testing, XGBoost
Machine Learning Scientist
2022 - 2022
Stealth Healthcare SaaS Startup
- Developed the ML model to predict the insurance claims status (approved or denied) using the claim documents.
- Analyzed the insurance claims data to find out insights and biases in the data for future model improvements.
- Made feature engineering to handle the categorical columns in the data.
Technologies: Artificial Intelligence (AI), Amazon Web Services (AWS), Machine Learning, Deep Learning, EDA, ETL, Data Science, Data Analysis, Data Analytics, Distributed Systems, GitHub, Analytics, Testing
Full-stack Machine Learning Scientist
2019 - 2022
PicsArt
- Developed the tag suggestion model to suggest tags to users on the shared screen. The first model was based on Word2Vec and then modified by Gensim, increasing the number of tags in uploaded images by 80%.
- Created the search ranking optimization algorithm, including new metrics creation and a re-ranking model based on Boosting algorithms. Increased the successful editing session after a search by 150%.
- Designed and developed an R&D environment for experimenting with search-related models on real search indexers and search results. The environment used a Python Dash library and SOLR search engine.
- Developed the content quality model, which predicts the future success of the image in the application, significantly increasing the discoverability and content diversity in the application.
- Oversaw the query expansion model, expanding the initial query with similar semantic queries. Significantly increased the number of returned results in search and content diversity.
- Increased the company's revenue by developing the model, which predicts show ads or offers screens to users.
- Developed recommender system models for stickers recommendations. The first model was based on user preferences and has shown more than 300% growth in stickers usage at each touchpoint when rolled out.
- Built a content discovery model based on learning reinforcement, exploring and exploiting newly created content.
- Managed a team of three members on different projects, contributing to their career and professional development.
- Developed a trend detection algorithm for the marketing and content teams, notifying them of any detected trend in the app.
Technologies: SQL, Python, PyTorch, Hadoop, PySpark, Recommendation Systems, Search, Rankings, Data Analytics, Machine Learning, Deep Learning, EDA, ETL, A/B Testing, Gensim, Natural Language Processing (NLP), Generative Pre-trained Transformers (GPT), Word2Vec, Research, Data Science, NumPy, Pandas, TensorFlow, Keras, Matplotlib, Seaborn, Financial Analysis, Scikit-learn, Amazon S3 (AWS S3), Amazon Web Services (AWS), Neural Networks, Deep Neural Networks (DNNs), Jupyter Notebook, SpaCy, Data Visualization, Business Intelligence (BI), Data Analysis, Data Reporting, Apache Spark, Model Development, Classification Algorithms, Statistical Modeling, Statistical Analysis, Distributed Systems, Ads, GitHub, Search Engines, Cloud Services, Testing, Statistical Methods
Data Scientist
2019 - 2021
PicsArt
- Worked on data sanity. The debugging finding and fixing bugs related to data events and data entities.
- Conducted global company analysis to find optimal ways of company development for the CEO. As a result, the main focus of the company's growth direction had been changed, and a new department was organized with the highest priority projects.
- Worked on A/B test experiment design and analyzed the results from the tests.
Technologies: A/B Testing, Big Data, Analysis, Databricks, Data Analysis, Data Analytics, Matplotlib, Seaborn, Statistics, Spark, PySpark, Python, Python 3, Time Series, Time Series Analysis, Trend Forecasting, Scikit-learn, Amazon S3 (AWS S3), Amazon Web Services (AWS), Neural Networks, Deep Neural Networks (DNNs), Jupyter Notebook, Search, Rankings, Data Visualization, Business Intelligence (BI), Data Reporting, Apache Spark, Model Development, Classification Algorithms, Statistical Modeling, Statistical Analysis, Distributed Systems, GitHub, Analytics
Software Engineer
2017 - 2019
ARLOOPA
- Developed an AR game with image recognition and tracking technologies.
- Tracked and fixed bugs in different projects mainly related to integrating AI components into AR apps.
- Handled a real-time sky segmentation project for the Arloopa app based on computer vision and image processing.
- Wrote new add-ons and plugins for existing applications and games.
Technologies: C#, C++, Python, Unity3D, Unreal Engine 4, Virtual Reality (VR), Augmented Reality (AR), Machine Learning, Deep Learning, Image Processing, Computer Vision, Scikit-learn, Jupyter Notebook, Back-end, Java, GitHub, Back-end Development
Machine Learning Scientist
2017 - 2018
Moscow State University Research Lab in Yerevan
- Worked on the time series analysis and prediction project, which uses recurrent neural networks to predict the future behavior of the time series.
- Worked on the stock market candle prediction using the empirical mode decomposition for time series.
- Researched new ways of analyzing and predicting time series.
Technologies: Artificial Intelligence (AI), Deep Learning, Machine Learning, Time Series, Time Series Analysis, TensorFlow, Keras, Python, Python 3, Financial Analysis, Scikit-learn, Neural Networks, Deep Neural Networks (DNNs), Jupyter Notebook, Data Visualization, Model Development
Experience
Insurance Claims Prediction
A model that uses patients' claims documents (EDI 835) and insurance claims (X12 837) and predicts whether or not the claims will be denied. The data columns are mostly categorical, and there is a lot of preprocessing and feature engineering to handle those kinds of features.
Preference-based Recommender System
The model captured user-to-item and item-to-item interactions to predict and recommend new items to current users. We trained the model on vast data from PicsArt's users and stickers usage. It is currently in production on all platforms and touchpoints.
Photo-based Recommender System
The model captured the best fit between photos and stickers based on previously selected, excellent quality photo edits. It used a metric learning approach to obtain the distance between a photo and sticker based on their visual characteristics.
Word Suggestion
The word suggestion model is used for tag suggestion and query expansion projects. Tag suggestion and query expansion use different data collection and preprocessing parts and the same training and inference parts. Used PySpark to collect the necessary dataset from the data lake and preprocess it for training and a modified version of the word2vec skip grams algorithm for training and inference.
Search Ranking Optimization
The project increases the relevance and diversity of search results in a platform. The project includes two sub-projects: new metrics creation and reranking algorithm based on the Boosting algorithm. The model relies on users' interactions during search sessions for the training process.
Ads Optimization
In this project, the model captured user preferences and interests to optimize ads in the application and show subscription offer screens based on the tools and content each user tends to like the most.
Trends Detection
The algorithm detects current trends and then creates a report with the most popular ones, similar tags, and other information. It uses the KL score, calculated using the probability of appearing tags and search queries during given dates.
Content Quality Predictor
The content quality prediction algorithm predicts the image quality using ResNet101 to get visual features and static information about the user and the published photo, considering, for example, the last popularity of photos posted by a user, country code of the photo or user, weight and height of the photo. The fully-connected neural network with binary classification was trained to predict photo quality. The scores returned by the model are used in the ranking formula to make the search results fresher, as new images have no information on interaction metrics to be used in the ranking formula.
Content Discovery
The exploit-explore algorithm is based on reinforcement learning through epsilon-greedy and Thompson-sampling and tries to optimally explore new content. This project uses content quality prediction algorithms to filter out bad quality images and applies epsilon-greedy algorithms only on good quality images.
Demand Prediction
The problem was to predict the demand of the used bonus points for clients and predict their behavior in 30-50 months. The model decreased the costs and increased the revenue of the clients. For solution was chosen the Mixture Models for Binomial, Log normal and Poison distributions.
Text-to-image Model Based on Stable Diffusion
The project involves the refinement and optimization of an existing text-to-image model using the Stable Diffusion framework. The project aims to enhance the model's performance through fine-tuning, making it more adept at converting textual inputs into visually appealing and contextually relevant images. This iterative process leverages the stability advantages of Stable Diffusion, ensuring a highly efficient and robust text-to-image synthesis model tailored to specific application needs.
Chatbot for Trade Companies
The project involves the creation of an advanced conversational AI system tailored specifically for trade companies. Leveraging state-of-the-art large language models (LLMs) and ChatGPT, the chatbot facilitates seamless and intelligent interactions, offering a personalized experience for users within the trade sector. This project aims to enhance communication, streamline customer inquiries, and provide valuable insights using cutting-edge natural language processing technologies, ultimately optimizing efficiency and customer satisfaction for trade companies.
Education
2016 - 2020
Bachelor's Degree in Informatics and Applied Mathematics
Lomonosov Moscow State University - Moscow, Russia
Skills
Libraries/APIs
PyTorch, PySpark, NumPy, Pandas, CatBoost, Matplotlib, Scikit-learn, LSTM, REST APIs, XGBoost, TensorFlow, Keras, SpaCy
Tools
Slack, PyCharm, Gensim, Seaborn, GitHub, Terminal, Claude, ChatGPT
Languages
Python, SQL, Python 3, Snowflake, C++, C#, Java
Frameworks
Spark, Apache Spark, Hadoop, Unity3D, Unreal Engine 4
Paradigms
ETL, Business Intelligence (BI), Testing, Model Context Protocol (MCP)
Platforms
Databricks, Jupyter Notebook, Amazon Web Services (AWS), Azure
Storage
Amazon S3 (AWS S3)
Industry Expertise
Social Media
Other
Machine Learning, Deep Learning, Recommendation Systems, Search, Rankings, Data Analytics, EDA, Natural Language Processing (NLP), Data Science, Data Analysis, Data-informed Recommendations, Big Data, Time Series, Time Series Analysis, Artificial Intelligence (AI), Trend Forecasting, Analysis, Recurrent Neural Networks (RNNs), Gated Recurrent Unit (GRU), Long Short-term Memory (LSTM), Neural Networks, Deep Neural Networks (DNNs), Data Visualization, Data Reporting, Model Development, Classification Algorithms, Distributed Systems, Ads, Analytics, Generative Pre-trained Transformers (GPT), Generative Artificial Intelligence (GenAI), Large Language Models (LLMs), Agentic RAG Systems, Prompt Engineering, Retrieval-augmented Generation (RAG), APIs, Agentic AI, OpenAI, Vector Databases, Workflow Automation, LangChain, Cloud Services, RAG Pipelines, Reliability, Design, Speech-to-Text (STT), AI Programming, AI Modeling, RAG Systems, Anthropic, Statistical Methods, Software Development, A/B Testing, Word2Vec, Research, Augmented Reality (AR), Feature Analysis, Statistics, Financial Analysis, Stock Price Analysis, Stock Market, Statistical Modeling, Statistical Analysis, Search Engines, Virtual Reality (VR), Image Processing, Computer Vision, Reinforcement Learning, Finance, Data Engineering, Back-end, Back-end Development, Stable Diffusion, Image Generation, Language Models, Generative Pre-trained Transformer 3 (GPT-3), OpenAI GPT-3 API, Llama 2, AI Chatbots, Chatbots, Social Networks
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring