JargonAI

JargonAI is a lightweight, end-to-end summarization pipeline for legal documents, built on Hugging Face’s FLAN-T5 model. It demonstrates data loading, preprocessing, model fine-tuning, and evaluation—all within a Colab-ready notebook.

Features

• Custom Legal Dataset
  Contains 60 human-written summaries of legal texts in Corpus_all.csv
• Pretrained Model
  Fine-tunes the google/flan-t5-small sequence-to-sequence model using PyTorch and the Transformers library.
• Data Preprocessing
  • Tokenization with T5 tokenizer
  • Train/validation split via sklearn.model_selection.train_test_split
  • Batching using PyTorch DataLoader
• Training Pipeline
  • Optimizer: torch.optim.AdamW
  • GPU acceleration (Colab GPU recommended)
  • Configurable hyperparameters (batch size, epochs)
• Evaluation Metrics
  Computes ROUGE and BERTScore to quantify summary quality.
• Interactive Notebook
  Colab-enabled Summa_rizzer.ipynb walks through each step, from dataset loading to metric visualization