Filip is a machine learning engineer with several years of professional experience. He's worked on large-scale problems at Amazon Web Services as a software developer and built natural language processing models as a research associate at the University of Zagreb. Filip's main interests are machine learning and natural language processing, with an emphasis on building text classification models.
Delivering data warehouse and ETL solutions as part of an Agile team using advanced ML techniques to improve performance and processes.
Helping build and improve infrastructure, application, and performance development and ensuring tight security including data encryption, security groups, and environment scanning.
Ensuring high-quality deliverables and implementing DevOps and security best practices in fast-paced environments.
Developed an LLM-based solution to determine which scientific articles are related to user-inputted free-text criteria.
Evaluated the LLM solution performance and demonstrated metrics proving considerable improvement over the previously implemented solution.
Worked with ML engineers to deploy solutions and define an optimal architecture for applying the LLM solution.
Developed prototype product recommenders showing customer purchasing patterns.
Built simple AWS Lambda functions to conduct an ETL workflow.
Worked with PySpark on large sets of data (>100GB of historical purchases).
Investigating, researching, and documenting caching methods in software.
Reproducing the most popular caching methods for predicting time-to-live from research papers.
Building a simulator and reinforcement learning model trying to solve TTL prediction for object caching.
Developed a Machine Learning sequence labeling model on text data that achieved above 0.9 F1 score.
Decreased inference time on a previously developed ML model without sacrificing their F1 score.
Used PySpark and Databricks to perform a large-scale data analysis that the company employed to drive future business decisions.
Developed multiple highly scalable Python web services that are currently serving production traffic.
Created a marketing slogan text generator using Hugging Face Transformers/text generation pipelines and customer-provided data.
Created a data ingestion and reporting process via multiple Google Cloud services: BigQuery, Cloud Functions, Cloud Endpoints, and Dataproc.
Ported existing R reporting code to a Python web service.
Developed a scraper to download a large (around 20,000) and diverse legal documents file (1,990 until today) from a European public repository.
Used ML to build a text classification model to automatically classify categories based on document content.
Created a dataset of legal documents and used it to train and evaluate the ML text classification model.
Shared results via Google Colab such that customers can interactively try the model performance with their held-out data.
Developed a search engine for Croatian legal documents.
Built a named entity recognition model in PyTorch by combining LSTM with a CRF.
Mentored several interns and wrote my master thesis on natural language processing.
Contributed to developing a scalable time-series database solution in Java and C++, which served around 1 million requests/second.
Served as the team scrum master and product owner.
Designed and implemented a network correlation engine microservice to handle networking events from the entire Amazon network (patent award https://patents.justia.com/inventor/filip-boltuzic).
Developed SQL reports to determine the promising retail strategies in a data warehouse.
Built an interactive tool in Java to speed up the processes in Oracle Data Integrator.
Developed small web applications for the accounting department using PL/SQL and Oracle Apex.
The project focused on predicting sale amounts based on historical data, previous sales, and regions.
Time-series forecasting was performed at both regional and global levels.
Experimentation included ARIMA and SARIMA models.
Developed a Django and Apache Solr web application with a microservices architecture.
The application stores and indexes documents within Solr.
The Django front-end handles user requests and communicates with Solr.
