CV
Education
- Doctor of Philosophy in Computer Science (ELLIS)Fall 2026 - PresentCWI & University of Amsterdam, Amsterdam, NetherlandsTRLab (CWI) & AMLab (UvA) — Advisors: Madelon Hulsebos & Jan-Willem Van de Meent. Focus: Multimodal Tabular Foundation Models.
- Masters of Science in Computer ScienceSeptember 2024 - PresentTufts University, Medford, MA, USAGPA: 3.95Expected Graduation: Spring 2026
- Masters of Science in Chemical EngineeringSeptember 2019 - December 2020Carnegie Mellon University, Pittsburgh, PA, USAGPA: 3.91
- Bachelor of Science with Honors in Chemical EngineeringSeptember 2016 - August 2019West Virginia University, Morgantown, WV, USACum Laude
Publications
- Synthetic Data Reveals Generalization Gaps in Correlated Multiple Instance Learning2025Ethan Harvey, Dennis Johan Loevlie, and Michael C. HughesML4H 2025 Symposium, Findings Track
- Demystifying the chemical ordering of multimetallic nanoparticles2023Dennis Johan Loevlie, Brenno Ferreira, and Giannis MpourmpakisAccounts of Chemical Research, 56(3):248–257, 2023
- Single Atom Alloys Segregation in the Presence of Ligands2023Salem, M., Loevlie, D. J., Mpourmpakis, G.The Journal of Physical Chemistry C, 127(46), 22790–22798, 2023
- Size-dependent shape distributions of platinum nanoparticles2022Ruikang Ding, Ingrid M. Padilla Espinosa, Dennis Loevlie, Soodabeh Azadehranjbar, Andrew J. Baker, Giannis Mpourmpakis, Ashlie Martini, and Tevis D. B. JacobsNanoscale Adv., 4:3978–3986, 2022
- Resolving electrocatalytic imprecision in atomically precise metal nanoclusters2022Anantha Venkataraman Nagarajan, Dennis Johan Loevlie, Michael J Cowan, and Giannis MpourmpakisCurrent Opinion in Chemical Engineering, 36:100784, 2022
Presentations
- Synthetic Data Reveals Generalization Gaps in Correlated Multiple Instance Learning2025Machine Learning for Health (ML4H) 2025 SymposiumSan Diego, CaliforniaPoster presentation
- Computer Vision for UAVs2023XChangeIdeas Pittsburgh
- Software Development for HER High-Throughput Experiments2020Carnegie Mellon University Chemical Engineering Masters Student Association Research Forum
- Mathematical Modeling and Optimization of an Ion Transport Membrane for Oxygen Separation from Air2018American Institute of Chemical Engineers National Research Conference, Computing and Process Control Division
Research Experience
- Tufts UniversityAugust 2024 - PresentGraduate Researcher with Dr. Michael HughesImproving the performance of deep learning models in situations with limited data quantity or quality.
- Using attention-based multiple instance learning (MIL) to predict precursors of dementia and stroke from 3D image data (MRI and CT).
- Published a paper at Machine Learning for Health on generalization gaps in correlated MIL.
- Identified and resolved a quadratic scaling bottleneck in the lab's MIL codebase, enabling experiments on larger datasets.
- Developing a regularization method to encourage more interpretable attention scores.
- Trained 3D CNN and Transformer models with multiple GPUs using the Tufts HPC.
- Implemented methods for supervised learning with noisy labels.
- Tufts UniversityJanuary 2025 - PresentGraduate Researcher with Dr. Jivko SinapovWork initially linked to Tufts Reinforcement Learning course.
- Used Group Relative Policy Optimization to improve Qwen2.5-Coder-7B's ability to generate SVGs from text descriptions. Drawing inspiration from recent works such as DeepSeek-R1 and AlphaMaze.
- Achieved an 18% improvement on a benchmark evaluating SVG aesthetics, alignment, and code validity.
- Currently working on understanding the limitations in vision-language models (i.e. Qwen3-VL) and open vocabulary instance segmentation models (i.e. SAM3) ability to count objects in dense scenes.
- University of Pittsburgh CANELaJune 2021 - January 2023Graduate Researcher with Dr. Giannis MpourmpakisApplied machine learning, Boltzmann statistics, and evolutionary optimization to predict material properties of metal nanoparticles.
- Contributed to neural architecture design, hyper-parameter optimization, and fair assessment of ML models on Salem et al.
- Proposed a novel method to initiate model weights from Yihao et al. that led to a 71% reduction in the RMSE on the datasets investigated in Loevlie et al.
- Wrote the ML applications and background section in the Nagarajan et al. review article.
- Collaborated with experimental research groups by using Boltzmann statistics to explain their findings in Ding et al.
- Carnegie Mellon UniversityDecember 2019 - December 2020Graduate Researcher with Dr. John KitchinDeveloped software tools to improve and automate experiment design and evaluation.
- Recreated image analysis tools in Python (originally in Mathematica) to be interactive, fast, and intuitive.
- Trained a convolutional neural network classifier to extract valuable information from experimental image data.
- Developed a Python package, nb_search, to efficiently sort through, locate and open Jupyter Notebook files.
- Regressed parameters and used them to cluster different bimetallic catalysts.
- West Virginia UniversityApril 2017 - August 2019Undergraduate Researcher with Dr. Fernando LimaMathematical modeling and non-linear optimization for chemical process design.
- Modeled, optimized, and economically evaluated a chemical process in MATLAB — Funded by the National Science Foundation.
- Presented findings at the American Institute of Chemical Engineers conference and was awarded second place in the poster competition.
Industry Experience
- KEF RoboticsJanuary 2023 - August 2024Senior Computer Vision and Machine Learning Engineer (2024), Computer Vision Engineer (2023)KEF Robotics is a Pittsburgh-based company that provides software only integrations enabling aerial autonomy on any unmanned aerial vehicle (UAV).
- Led a team of five engineers on a one-year, $500K project where I was responsible for task breakdown, budgeting, and advanced ML research and implementation.
- Led the development of efficient on-device object detection, monocular depth prediction, and 3D map generation from monocular camera images. Showcased these capabilities at two in-person demos.
- Enhanced hazard detection for UAVs with Mask2Former, a transformer-based universal image segmentation model. Fine-tuned the model to segment a new class (power lines) and generalize to a new image modality (infra-red) using transfer learning.
- Optimized our image segmentation neural network architecture, resulting in a significant 45% boost in inference speed with only a 1% loss in accuracy.
- AiThEliteDecember 2020 - May 2021Lead Data ScientistAiThElite is a Pittsburgh-based startup company using AI to improve the college athlete transfer process.
- Developed web scraping scripts using Beautifulsoup and Selenium to automate data retrieval and updating.
- Developed and automated the feature engineering with Numpy and Pandas.
- Applied machine learning algorithms using Numpy and SkLearn to generate intelligent predictions and insights from the data.
- Built the frontend and backend of the AithELITE EliteAI website with Django, hosted on AWS.
Internships
- ContentsPalMay 2025 - August 2025Multimodal AI InternMIT professor led AI startup in the insurance space.
- Investigated and implemented learning-based duplicate detection and open-vocabulary instance segmentation.
- Worked with React and React Native to test new and existing features.
Projects
- GPT4ReadabilitySummer 2023Natural Language Processing, Deep Learning, Open-Source
- Developed a Command Line Interface (CLI) that leverages large language models (LLMs) and vector databases with LangChain and llama.cpp to generate a comprehensive README file and suggest code improvements for any GitHub repository.
- Supports running with cloud-based LLMs or running locally with open-source LLMs.
- Supports 15 different programming languages.
- SkinsAIFall 2022Computer Vision, Deep Learning, Hosted
- Developed a free-access, diagnosis tool for classifying moles as benign or malignant.
- The convolutional neural network classification model was written, trained, and evaluated using PyTorch.
Technical Skills
Languages
- Python
- MATLAB
- Java
- JavaScript
- C++
ML & AI
- PyTorch
- TensorFlow
- Scikit-learn
- Computer Vision
- Transformers
- LLMs
- Reinforcement Learning
- Graph Neural Networks
- Optimization
Computing
- NumPy
- Pandas
- SciPy
- Matplotlib
- HPC Clusters
- TensorRT
Awards
- 2026First place out of 10 teams in the BrainStorm Neural Decoder Challenge, real-time auditory decoding from 1024-channel ECoG; 95% accuracy, sub-ms inference, edge-deployable model.
- 2024Awarded Community Grant from Hugging Face to demonstrate Depth Anything results on videos.
- 20222nd place out of 24 teams in The Pitt Challenge Hackathon for building SkinsAI.
- 20203rd place in the Chemical Engineering Masters Student Association Research Forum, Poster Competition.
- 2020Category winner in The Pitt Challenge Hackathon "Largest impact on healthcare workers" category.
- 20191st place in AVEVA's National Simulation Competition (advanced category).
- 20182nd place in the American Institute of Chemical Engineers National Poster Competition, Computing and Process Control Division.
Community Involvement
- Youth robotics team working on tools for blind soccer players2024Industry Volunteer
- Organized a profit sharing event to raise funds for the flooding in Pakistan2022Leadership
- Volunteered at an outreach event to help encourage students to pursue STEM2022Leadership
- Volunteered to conduct science experiments with elementary students2021STEM Education