EY Aleph: deep learning applied to jurimetrics practice
summary
Collects public court records from 200+ Brazilian tribunals, scrapes PDFs (handles CAPTCHAs with a small CNN/vision model), converts PDFs to text, and builds a human-labeled corpus using law students to tag phrases. Implements a pipeline on cloud storage + Celery, Rails/Ember front end, Jenkins CI and Slack monitoring. Trains phrase classifiers and a decision classifier (granted/overruled) using bi-directional LSTM with task-specific embeddings, replaces brittle regexes, and makes legal-team workflows roughly 10x faster while enabling future predictive provisioning models.
AI-written summary · about
chapters (5)
related talks
01h 24mThe use of AI by the Judiciary in LatAm Countries Hallucination or Opportunity?
44mExtracting Structured Data from Legal Documents
39mEnterprise-ready document processing for AI
33mDeep research at scale
43mGetting started with AI for free
31mTaking AI/ML to the next level: Text!
40mJudgement Day: Benchmarking "Black Box" LLMs With Open Legal Datasets
33mUnlocking Insights From Multimodal PDFs Using OpenSearch and Vision-Language Models