Tobivictor/legal_datasets

Legal Case Documents Dataset Snapshot for GaiaNet

This repository contains a Qdrant database snapshot of vectorized legal case documents from Nigeria and the UK. The data has been processed and converted into embeddings suitable for use as a knowledge base in a Retrieval-Augmented Generation (RAG) system.

Purpose

This dataset was created to serve as a specialized knowledge base for the GaiaNet ecosystem. The goal is to enable a Large Language Model (LLM) to answer questions and provide information based on the contents of these legal documents.

Dataset Creation Workflow

The snapshot was generated using the following process:

1. Source Data: The original data was a JSON file containing pairs of instruction (case metadata) and response (full case text).

2. Pre-processing: The raw text was processed by a Python script (pre_chunker_v2.py) to split very long documents into smaller, manageable chunks. This was done to accommodate the AI model's context window and to prevent memory overload. Chunks were split first by paragraphs, and then by sentences if a paragraph was still too large.

3. Vectorization Toolchain:

   • Embedding Model: gaianet/Nomic-embed-text-v1.5-Embedding-GGUF
   • Vector Database: Qdrant
   • Runtime: WasmEdge
   • Embedding Tool: paragraph_embed.wasm from GaiaNet's embedding tools.

How to Use This Snapshot

1. Download the legal_cases_snapshot_PARTIAL_2_percent.tar.gz file from this repository.
2. Decompress the archive. This will produce a directory named default.

   tar -xzvf legal_cases_snapshot_PARTIAL_2_percent.tar.gz
   

3. Place the default directory inside the Qdrant snapshots volume that your GaiaNet node or Qdrant instance is using.
4. The GaiaNet node should be configured to load the snapshot from this collection.

Dataset Structure

• The .tar.gz file contains a single directory named default.
• This directory holds the Qdrant snapshot files for the collection.
• Each point in the collection consists of:
  • A 768-dimension vector representing the text chunk.
  • A payload containing the original source text of the chunk.

---

This dataset was prepared by Tobivictor. Date: September 15, 2025