Experimental Nuclear Physics, Artificial Intelligence,
and Machine Learning Specialist
Research Scientist
Welcome! I’m Dr. Mariana Khachatryan, a Research Scientist in experimental nuclear and particle physics based in Virginia’s Hampton Roads region, originally from Yerevan, Armenia. My work lies at the intersection of experimental nuclear physics and artificial intelligence, where I apply machine learning techniques to extract physical insight from complex, high dimensional datasets. With more than ten years of experience working on experiments at Thomas Jefferson National Accelerator Facility, I specialize in the analysis of large scale particle physics data from CLAS and GlueX collaborations, including studies of meson photoproduction with exotic quantum numbers and 3D proton structure via deeply virtual Compton scattering (DVCS). My research combines statistical inference, Monte Carlo methods, Machine Learning and Artificial Intelligence to uncover the structure of the nucleon and interpret experimental measurements. Growing up in a family of scientists, I was naturally drawn to discovery and problem solving. During my undergraduate studies at Yerevan State University, I attended the DESY Summer School in Germany, where I completed my first data analysis project and presented results to both technical and non technical audiences. This experience sparked my long term interest in using computational methods to understand physical systems. I earned my Ph.D. in Physics from Old Dominion University in 2019, following research at Jefferson Lab, and continued as a postdoctoral researcher at Florida International University. My work has resulted in 40+ peer reviewed publications, including research featured in Nature, and has been recognized by multiple institutions. In parallel with my experimental research, I have developed strong expertise in machine learning and artificial intelligence, including neural networks, deep learning, and optimization techniques. My recent work focuses on accelerating DVCS Compton Form Factor fitting using PyTorch based implementations, enabling efficient batch evaluation of custom loss functions and integrating modern ML approaches into physics analysis workflows. I am particularly interested in advancing the use of AI and machine learning in fundamental physics, as well as applying these methods to other complex scientific and technological problems. Outside of research, I enjoy traveling, gardening, and gaming on Nintendo Switch activities that keep me curious and creative.
My qualifications span a range of tools and specializations that strengthen my expertise. Key highlights include:
Here you can see some of the projects that I have worked on throughout my career.
Feb 2026 – Present · Old Dominion University Research Foundation
Using the PARTONS software framework dedicated to the phenomenology of 3D hadron structure, parametrizing DVCS CFFs via ANNs by implementing ANN training on available global data from DVCS observable measurements.
Skills: C++, Artificial Neural Networks
Feb 2026 – Apr 2026 · Old Dominion University Research Foundation
Implemented batch calculation of a custom loss function in DVCS CFF fitting in the Gepard Python package using PyTorch tensor operations for data points sharing the same physics formulas, resulting in a 30x speedup in ANN training.
Skills: Python, PyTorch, Artificial Neural Networks
Used Python and C++ for exploratory data analysis, data engineering and predictive analysis of particle physics data at Jefferson Lab. Developed data analysis strategy and timeline. Used Least Squares and unbinned Likelihood statistical methods to clean high-volume data via Probabilistic Event Weightings. C++ to classify particles and identify exotic mesons via Unbinned Maximum Likelihood fitting. Used bootstrapping for error estimation. Delivered presentations to technical and non-technical audiences. Mentored team members.
Led a team of PhD researchers to develop supervised learning models (Linear Regression, SVM, XGBoost) for predicting car prices using structured automotive data. Used Exploratory Data Analysis and Feature Engineering to improve model performance, achieving R² of 0.88. Applied SHAP values to identify key features driving model prediction. Delivered technical findings and business implications to product stakeholders from partner companies (Carmax, Upstart, etc.).
Developed time series forecasting models using LSTM networks and Linear Regression to predict clothing sales trends. Engineered features to describe time dependent and serially dependent properties of Time Series to enhance predictive power of Linear Regression model. Achieved R² score of 0.91, demonstrating high accuracy on real-world retail data. Compared model performances to establish tradeoffs between interpretability and predictive strength.
Have used OpenAI API and LangChain framework for developing applications powered by large language models (LLMs), to develope a PDF chatbot that uses natural language processing (NLP) to understand queries and OpenAI API to extract email address from provided PDF resume.
Fine-tuned model for text classsification using the OpenAI API and served with FastAPI.
Used convolution neural network (CNN) from Keras/Tensorflow to classify images with/without a face mask to 97% accuracy.
Performed text processing and classification using NLP techniques, leveraging TensorFlow LSTMs and a Naive Bayes model.
Generated images of handwritten digits using Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs).
Have used K-Means clustering unsupervised learning algorithm to reduce number of colors in sailboat image to three.
Comparison between K-Means and DBSCAN on moon shape clusters. I have apply DBSCAN on clusters with irregular shape and have shown that it gives better performance than K-Means clustering.
Please contact me if you have any questions about my skills, previous work experience, or any projects I have completed. I am always open for conversations with recruiters about interesting career opportunities and would love to hear about what is available.