Armen A. Agalyan
← about

Work

Machine learning engineer focused on empirical AI research, scalable ML systems, and production deployment. Currently consulting at Magical Labs.


experience

Magical Labs

Machine Learning Engineer · Oct 2022 – Present

Consulting on production ML systems across a range of client domains. Built a steganography system achieving 96.8% bit recovery under JPEG compression, embedding hidden messages in Stable Diffusion images via DDIM noise modulation. Deployed a recommendation system on AWS OpenSearch + Celery for a 200k MAU mobile games marketplace, increasing revenue-per-click by 23%. Led a predictive analytics dashboard for a $10B real estate portfolio that cut monthly reporting time by 95%.

Stealth Startup

Research Engineer · Oct 2024 – Jul 2025

Fine-tuned GPT-2 for generative percussion modeling using FSDP and PEFT for professional music production. Engineered a high-performance C++ MIDI tokenizer and real-time inference pipeline with ONNX Runtime, achieving a 10× CPU speed-up across major DAWs.

Media Instinct Group

Machine Learning Specialist · May 2021 – Mar 2022

Reduced Spark ETL job execution time by 67% (6 → 2 hours) by optimizing execution plans at scale. Built an ARIMA time series model forecasting GRP for commercial media buying, directly informing strategy for 50+ corporate clients.

MaximaTelecom

Data Scientist · Feb 2019 – May 2021

Built an NLP pipeline classifying unstructured text into 80+ categories for an expense management service. Engineered an XGBoost pipeline deriving socio-demographic signals from raw data for 5M+ users, improving prediction accuracy by 12%.


education

Lomonosov State University

MS, Applied Mathematics · 2020

Research in extreme value theory — rate of convergence in limit theorems for maxima of random variables, conducted in the Laboratory of Probability.


skills

Languages

Python, C++, Scala, SQL, Bash

Frameworks

PyTorch, Keras, FSDP, Hugging Face, ONNX Runtime, Scikit-learn, XGBoost

Infrastructure

AWS, Apache Spark, Airflow, Docker, MLflow, Celery, Elasticsearch, Postgres