Building a Domestic Risk Assessment Tool: Data Cleaning, Risk Scoring Models and Privacy Compliance | Sabina Firtala
Listen to or watch on your favorite platform
Show Notes
How do you build an accurate, privacy-compliant domestic risk assessment tool that frontline teams can actually use? In this episode Sabina Firtala — who leads Frontline's AI product development and brings experience in data wrangling, model validation, and applied analytics from finance, SaaS, and mission-driven projects — walks through a practical roadmap.
We cover problem framing and project scope; sources like case management, public records, and surveys; and hands-on data work: cleaning, linking, and feature engineering. Sabina explains risk scoring approaches and model architecture, evaluation metrics and bias assessment, plus privacy, ethical considerations, and legal data governance. You’ll also hear about deployment into frontline workflows, user interface and decision-support design, training and stakeholder trust, ongoing monitoring and drift detection, and examples of impact on triage and resource allocation. The conversation closes with collaboration strategies, funding and scaling, open documentation for reproducibility, and concrete lessons learned.
Listen for actionable guidance on data cleaning, building and validating risk scoring models, and ensuring privacy compliance so you can design responsible, usable domestic risk assessment tools.
Resources Mentioned
Tools, books, papers, and other resources mentioned in this episode
- company Frontline 100 Guest's company applying AI to domestic abuse risk assessment
- tool Lizzy App-based risk assessment tool replacing paper surveys for police and social workers
- tool MLflow Experiment tracking in the guest's MLOps workflow
- tool PyCaret Tried for modeling but dropped as too heavy for the app
- course Hyperskill JetBrains-backed Python learning platform the guest highly recommends
- course Ironhack Bootcamp where the guest studied data analytics
- other Our World in Data Followed for reports and visualizations; even a job lead
- community The Portfolio Collective Networking site for multi-skilled freelancers with social events
- book Look What You Made Me Do Recommended reading on domestic abuse, on guest's list
- other Maid Netflix show depicting domestic abuse through a woman's lens
- other Baby Reindeer Netflix show depicting abuse through a man's lens
- tool XGBoost Boosting model tried before logistic regression performed best
- community Kaggle Survey-data competition where linear models won
This list is generated from the episode transcript with AI and may contain inaccuracies. If you notice something wrong, let us know.
Timestamps
Timestamps coming soon...