- Engineered a time series model that optimally chooses the best-fit call center agent for a transition between roles. Utilizing the agent’s KPIs, the model identifies correct agent fitness in 75% of the cases, 38% more than the human counterpart. Accurate forecasting allows better planning for seasonal shifts in the company. - Engineered a model that forecasts Small Business segment locations by analyzing local market dynamics. With a 1% conversion rate, the solution allows to gain >$1M MRR by successfully identifying the business sector. - Finetuned BERT-like model for a multi-output classification where the model reads the incoming emails and classifies them into buckets for future processing. The model saves human agents >4 minutes of reading per email.- Researched internal support ticket systems, to identify cost-cutting business opportunities. Through data analysis, and insights, steered the team into an opportunity that saves >$ 200,000 a year through automation. Optimized a critical to the task data processing pipeline, and decreased the execution time from 3 days down to 10 minutes on each run. Stack: GCP, PyTorch, Pandas, SkLearn, Python, SQL, git, JupyterLab

- Developed a proprietary deep learning ensemble algorithm that beats state-of-the-art accuracy by 5% in drug-target interaction prediction.- Conducted state-of-the-art machine learning research in the biomedical field focused on modeling interactions between FDA-approved drugs and human proteins.- Developed a data pipeline with a proprietary dataset of interactions composed of millions of drug-target interaction data points.Stack: Google Cloud, PyTorch, Pandas, SkLearn, Python, SQL, git, JupyterLab

- Wrote critical asynchronous multithreaded code for an internal message-passing system to resolve race conditions.- Wrote automated deployment scripts in PowerShell for the Azure Cloud health monitoring service. - Augmented API capabilities and security.