Portfolio - Dragos Pop

About me

MSc Data Science graduate with six years of experience using data to solve real-world problems and helping businesses become more profitable. On my path bridging Amsterdam, Singapore, and Melbourne, I worked with cutting-edge AI and Data Engineering tools to drive impact for leaders in renewable energy retailing, transportation, and healthcare.

Projects I have delivered include: data pipeline orchestration, CI/CD workflow, music generation (GenAI), shipment classification, sales forecasting, system automation, duplicates identification, berth scheduling, recommender systems, model evaluation, dashboard design, price and rating prediction, etc.

Portfolio (inverse chronological order)

Invoicing Pipeline Lite

Aug 2026

Built a Dagster pipeline that reads monthly contractor hours from an Excel workbook, generates a branded PDF invoice, and emails it directly to the client

Grid Capacity Pipeline

Aug 2025 ~ Flo Energy

Automated a pipeline that fetches a byte-encoded PDF report via API, parses multiple tables, loads them into Snowflake via Amazon S3, and refreshes dbt models powering a Holistics dashboard

Selective Dagster Code Location Deployment

Aug 2024 ~ Flo Energy

Expedited Dagster branch deployment CI through a GitHub Actions job which dynamically selects code locations for rollout based on PR labels

Generating Music from Raw MIDI

Aug 2023 ~ Network Institute VU

Developed and trained GPT-like transformers on HPC to generate music from Raw MIDI

Goodreads Book Rating Prediction

Apr 2023 ~ Kaggle competition

Predicted the review ratings of books on Goodreads, scoring an F1 score five times higher than the naïve baseline

Ranking Hotel Listings on Expedia

May 2022 ~ MSc Course "Data Mining Techniques"

Ranked hotel listings by their likelihood of being booked, obtaining an NDCG@5 over two times higher than the random baseline

Classifying Held Shipments

Jan 2022 ~ FedEx Express

Developed a LightGBM model for classifying whether a shipment will be held awaiting correct information, outperforming the deployed neural network by 53%