Attus Procuradoria Digital

Data Scientist

• Develop classification and regression models to predict: probability of success, type of decision, and case behavior throughout proceedings;
• Design and evaluate different approaches: classical models (Logistic Regression, XGBoost, Random Forest), NLP-based models, embeddings, Transformers, and hybrid models (statistical + textual semantics);
• Analyze initial petitions and other legal documents to extract: legal entities (NER), themes, grounds, claims, relevant attributes, and similarities between cases;
• Create semantic and structured features to feed the models;
• Stay current with state-of-the-art methodologies and algorithms to optimize accuracy and other relevant metrics in the context of predicting legal decisions;
• Perform deep exploratory data analysis (EDA) on legal datasets provided by the system;
• Test different data structures to improve model performance;
• Build proofs of concept (POCs) with LLMs focused on: summarization of filings, case progress and decisions; assisted generation of legal texts (drafts, preliminary opinions, etc.); support for attorneys with specialized agents;
• Work with RAG (Retrieval-Augmented Generation) connected to legal knowledgebases; use semantic search to capture legal understanding; fine-tune or instruct generative models for specific tasks and benchmark open-source and proprietary models;
• Validate models using robust metrics (AUC, F1, KS, etc.);
• Apply explainability and feature-importance techniques to interpret decision drivers;
• Produce technical reports explaining: methodology, interpretation of results, relevance of variables, limitations and next steps; document versions, experiments and training procedures;
• Specify technical requirements for integrating models;
• Deliver artifacts and technical guidance on model input/output.