Publications
2025
2023
2022
2020
Projects
Tools
Urartu
ML pipeline framework with smart caching and remote execution. Chain actions into workflows, run anywhere — local machines to HPC clusters. YAML-configured with built-in experiment tracking and SLURM integration.
OrganizeNoc
Browser extension suite for researchers to manage academic papers library. Features metadata extraction, highlight processing, citation BibTeX export, and AI-powered querying.
Aim
Open-source experiment tracking tool designed with flexibility for machine learning workflows. Facilitates detailed monitoring and comparison of experiments.
tmynNLP
Natural language processing pipeline equipped with mainstream abstractions. Supports a diverse array of NLP tasks.
Teaching
Introduction to LLMs - Lecture (Notebooks)
Deep Learning - Lecture (Notebooks)
Experiment Tracking - Workshop
Essentials of experiment tracking in ML research, with an in-depth exploration of Aim and a practical demonstration integrating it into an NMT fine-tuning pipeline.
LxMLS Guide - Summer School
Lisbon Machine Learning Summer School lab guide with setup instructions, toolkit walkthroughs, and hands-on exercises.
PyTorch Optimization - Workshop
Strategies for efficient dataloader usage, parallel computation, operator fusing, and more to enhance PyTorch code performance.
Resume (download)
Education
Ph.D, Computer Science, Technical University of Darmstadt
M.Sc., Computer Science, American University of Armenia
B.Sc., Nuclear Physics, Yerevan State University
Work Experience
Applied Research Intern, NVIDIA, San Francisco Bay Area, CA
- Research on improving Reasoning Language Models capabilities on STEM domain.
Applied Research Intern, NVIDIA, Yerevan, Armenia
- Improved code-reasoning performance of open-source LLMs on SWE-Bench through targeted evaluation, error analysis, and iterative modeling iterations.
- Built a localization system Artsiv achieving state-of-the-art results on SWE-Bench (1 publication).
Doctoral Researcher, Technical University of Darmstadt, Darmstadt, Germany
- Conducted mechanistic interpretability research to predict pre-generation instabilities and hallucinations (factual self-unawareness) at inference time in LLMs (2 publications).
- Created multi-agent system to generate human-chatbot interaction dialogues at scale (1 publication).
- Developed Urartu, a lightweight slurm submission + reproducibility framework for faster iteration.
Machine Learning Researcher, YerevaNN, Yerevan, Armenia
- Investigated interpretability of the Chameleon model (FAIR, Meta AI collaboration) (1 publication).
- Researched transformer alternatives for text generation: diffusion models, mixture of expert models, and language model bootstrapping (FAIR, Meta AI collaboration).
- Trained a molecular representation learning model (FAIR, Meta AI collaboration) (1 publication).
- Experimented on cross-lingual zero-shot transfer using prompt-based learning and adapters.
- Researched bilingual sentence alignment in low-resource setups (Master’s Thesis).
- Built a machine translation system for the WMT20 Biomedical domain tasks (1 publication).
Machine Learning Engineer, AimStack, Yerevan, Armenia
- Integrated the Aim experiment tracking tool with popular ML frameworks and MLOps tools.
- Authored technical blog posts to increase community awareness and adoption (3 blog posts).
Machine Learning Engineer, ZERØ, Yerevan, Armenia
- Researched and developed a legal document classification system.
- Built tmynNLP, a framework used to develop a multiclass document classification pipeline.
Software Engineer, Birthright Armenia, Yerevan, Armenia
- Led and built a large-scale CMS system with email management system.
Physics Researcher, CANDLE SRI, Yerevan, Armenia
- Measured electron-beam emittance on the AREAL linac (1 publication).
- Built a tooling suite for accelerator control and beam diagnostics.
Web Developer, Floopen, Yerevan, Armenia
- Developed and maintained a general-purpose content management system.
Teaching Experience
Teaching Associate, Technical University of Darmstadt, Darmstadt, Germany
- Taught the Introduction to LLMs course as a Teaching Associate.
- Conducted workshops on PyTorch optimization and experiment tracking and reproducibility.
- Supervised 4 M.Sc. theses in mechanistic interpretability.
Core Monitor, Instituto Superior Técnico – LxMLS 2023, Lisbon, Portugal
- Authored a step-by-step guide explaining the transformer architecture.
- Facilitated hands-on lab sessions for researchers, graduate students, and industry professionals.
Teaching Associate, American University of Armenia, Yerevan, Armenia
- Deep Learning: Taught natural language processing lectures and led problem-solving sessions.
- Data Structures and Algorithms: Led problem-solving sessions and created homework assignments.
Skills
Computer Science: Algorithms Design, Data Structures
Languages: C, Python, CUDA, JavaScript
Machine Learning: PyTorch, xFormers, FairSeq/MetaSeq, AllenNLP, PyTorch Lightning
Database Systems: SQL, NoSQL, GraphQL