Loubensky B.

Key Skills

Software

Other

Don't disclose

Top Subject Matter

Document OCR and structured data extraction for policy research

Program analytics automation using generative AI tools

Top Data Types

Document

Top Task Types

Text Generation

Transcription

Freelancer Overview

My experience preparing and structuring data for machine learning models comes primarily from my work at the Open Justice Lab, where I designed a Python based OCR pipeline to convert over 5,000 heterogeneous FOIA PDF reports into clean, structured CSVs. This required building extraction scripts that enforced consistent naming and merging rules across inconsistent source formats, a process that mirrors the rigor and attention to detail required in data labeling and quality assurance for AI training datasets. The result was a reliable, multi-state dataset of over 100,000 records that downstream researchers could trust for policy and mortality analysis, which reinforced for me how much model performance depends on the quality of the data feeding it. I have also worked directly on the training data side of machine learning pipelines. On my Health Event Anomaly Detection Platform, I used GAN augmented data to expand and diversify the training set for an XGBoost classifier, which improved outbreak monitoring accuracy by 24 percent. Combined with my coursework and hands on experience across the full data lifecycle, from raw ingestion in Kafka and Spark to structured storage in PostgreSQL, I bring both a technical understanding of how training data shapes model outcomes and a demonstrated ability to build the careful, standardized processes that produce trustworthy datasets at scale.

Labeling Experience

Boston University Innovate@BU — Automations and Data Solutions Intern (Sept. 2025–Present)

Don't discloseDocumentText Generation

Created engagement reporting data systems and automated workflows to integrate third-party platforms for program process automation. Used generative AI tooling as part of workflow creation to support program operations and data integrations. Designed dashboard outputs to track participation, outcomes, and program performance over time. • Built 15+ automated workflows using Zapier, Airtable Automations, and generative AI tools • Integrated CRM and calendaring platforms to automate program processes • Developed engagement reports and dashboards for tracking participation and outcomes • Improved accuracy and reliability of 1,000+ students engagement data via a new cloud database system

2025 - Present

Data Science Intern - Open Justice Lab

DocumentTranscription

As a Data Science Intern, you supported a faculty research client team by building multi-state datasets for prison overcrowding analysis. You designed a Python-based OCR pipeline to transform heterogeneous FOIA PDF documents into structured CSV outputs with standardized naming and merging rules. You delivered client-ready outputs including cleaned datasets, reproducible notebooks, and a written report synthesizing findings and recommendations. • Build multi-state datasets (100,000+ records) to enable policy and mortality analysis across priority states • Design and implement a Python OCR pipeline to convert FOIA PDFs into structured CSVs • Produce cleaned data packages, reproducible notebooks, and written reports for clients • Standardize extraction layouts and enforce consistent naming and merging rules

2025 - 2025

Education

Work History