Titulo Estágio
Explainability for Machine Learning Models in Financial Decision Systems
Local do Estágio
Remote, Hybrid or Lisbon/Porto/Coimbra office
Enquadramento
Feedzai is a global leader in risk management software using machine learning for real-time financial crime prevention. As part of our commitment to Responsible AI, we are developing next-generation explainability tools that go beyond standard attribution techniques, to support real analyst workflows and governance standards. This project is anchored in Feedzai’s Responsible AI group, working to ensure that all models served within our ML infrastructure are explainable, with behavior that is consistent, efficient, and fair across model types.
The internship is part of an initiative to develop a unified backend solution, providing intuitive, model-agnostic explanations that protect proprietary feature definitions while enabling analysts to reason about risk, using historical data patterns. This internship will provide an opportunity to engage in applied research, on topics such as efficient SHAP-like methods, surrogate and glassbox modeling frameworks, or semantic concept attributions,
Objetivo
The main objective of this internship is to explore, prototype, and help productize efficient explainability solutions for ML models in fraud detection workflows. Key goals include:
1. Algorithmic Innovation: Investigate ways to make local feature attribution practical (e.g., SHAP approximations, amortisation, surrogate models).
2. Design of Concept-Based Explanations: Research methods to map feature attributions to semantically meaningful concepts aligned with analyst reasoning.
3. Prototype Development: Design and evaluate lightweight backend solutions that deliver explainability with low latency and memory usage.
4. UX Alignment: Collaborate on integration strategies with our Frontend solution, focusing on data visualisation and transparency.
5. Scientific Contributions: Support the development of scientific outputs, and potentially co-author an invention disclosure or patent application.
Plano de Trabalhos - Semestre 1
The student will engage in exploration, background study, and ideation to prepare for hands-on work.
1. Orientation and Research Immersion: Introduction to Feedzai's Responsible AI initiative and existing explainability tools.
2. Literature Review: Study of current methods in local explainability (SHAP, LIME, glassbox models).
3. Initial Explorations and Research Planning: Identify experimental directions under supervision, to prototype in the second semester. Define an actionable research question and metrics for evaluation
Plano de Trabalhos - Semestre 2
Full-time engagement and deep technical involvement in the research and development of an explainability solution.
1. Prototype Development: Implement and test backend components for attribution computation, caching and concept retrieval.
2. Evaluation and Refinement: Benchmark latency, consistency, and analyst performance metrics across models and experimental trials.
3. Scientific Output and Knowledge Transfer: Contribute to scientific papers (e.g., on efficient SHAP or attribution conceptualization). Co-author internal reports or patent drafts where applicable. Present findings to stakeholders within the Responsible AI group.
Condições
Interns may work remotely or from Feedzai’s offices in Lisbon, Porto, or Coimbra (hybrid models supported). We offer flexible working hours.
The intern will receive dedicated mentorship by senior AI researchers, with regular one-on-one check-ins and weekly team syncs. Access to internal ML platforms, compute infrastructure, and proprietary datasets will be offered.
All scientific contributions will be acknowledged in patents and publications where applicable.
Observações
The internship offers exceptional exposure to real-world problems and constraints in the FinTech domain, with a strong emphasis on AI for good and responsible AI practices.
Candidates with interests in ML research, efficient computing, and responsible AI design are especially encouraged to apply. Past interns in this team have transitioned into research scientist roles or pursued PhDs with joint supervision from Feedzai staff.
Orientador
Iker Perez, Inês Oliveira e Silva
iker.perez@feedzai.com, ines.silva@feedzai.co 📩