Generating Music from Raw MIDI
Aug 2023 ~ Network Institute VU
Developed and trained GPT-like transformers on HPC to generate music from Raw MIDI
MSc Data Science graduate with four years of experience using data to solve real-world
problems and helping businesses become more profitable.
Projects I have worked on include: pipeline orchestration, music generation (GenAI), shipment
classification, sales forecasting, system automation, duplicates identification, berth
scheduling, recommender systems, model evaluation, dashboard design, price and rating
prediction, sentiment analysis, etc.
Developed and trained GPT-like transformers on HPC to generate music from Raw MIDI
Forecasted daily sales for 28 days in the future of 823 products sold in a Walmart store
With the release of the Hypertension Plus app, OMRON Healthcare wanted to raise the engagement among users by keeping track of specific groups of patients, such as the ones who have not accessed their account in the last week or the ones who have not completed their blood pressure measurements, and targeting them with email reminders
Port of Amsterdam demanded to know if giving up the current First Come, First Served strategy of assigning vessels to berths would produce any benefits, especially after the new sea lock will be ready and (more) larger ships will be allowed to moor in the harbor
Orchestrated a pipeline that retrieves the latest exchange rates through an API, loads them first into an Amazon S3 bucket and then into a Snowflake table
Engineered a web application that reads text and outputs an event in .ics format, using Streamlit, LangChain, and OpenAI API
Developed and trained GPT-like transformers on HPC to generate music from Raw MIDI
Applied classical Machine Learning techniques to data collected from smartphone sensors
Predicted the review ratings of books on Goodreads
Investigated the decentralization degree and trading patterns of the richest BTC, DOGE, and ETH holders
Ranked hotel listings based on their likelihood of being booked, obtaining an NDCG@5 over two times as good as the random baseline ranking
Designed and implemented a scalable Machine Learning pipeline that identifies duplicate entries in a bibliography with 82% accuracy
Enhanced the feature space for fire effect modeling through web scraping and designed a dashboard to visualize the results
Developed a LightGBM model for classifying whether a shipment will be caged (secured until the correct information is obtained)
Forecasted daily sales for 28 days in the future of 823 products sold in a Walmart store
Programmed a class in OOP fashion that contains several methods for evaluating the performance of Machine Learning models
Explored, pre-processed, and combined UN speeches with Life Ladder data to perform sentiment analysis, predict a country's happiness index, and classify the region of a speech
Built a multivariate, multistep, single-output LSTM that predicts the following two weekly averages of systolic measurements of active OMRON connect users
Analyzed data to gather insights about the OMRON devices, the OMRON connect app, and its users' blood pressure
Hypertension Plus is an app meant to reduce the friction between patients with hypertension and their physicians with the help of blood pressure home monitoring. To measure the performance of the app and gather actionable insights, a dashboard depicting critical metrics was designed
With the release of the Hypertension Plus app, OMRON Healthcare wanted to raise the engagement among users by keeping track of specific groups of patients, such as the ones who have not accessed their account in the last week or the ones who have not completed their blood pressure measurements, and targeting them with email reminders
Port of Amsterdam demanded to know if giving up the current First Come, First Served strategy of assigning vessels to berths would produce any benefits, especially after the new sea lock will be ready and (more) larger ships will be allowed to moor in the harbor
Inspecting and modeling the spread of epidemics based on various vaccination strategies
Investigated the trade-off between money saved by removing beds and the decrease in the hospital's rejection rate