Israeli Supreme Court Dataset — Web Platform

A self-hosted web platform for exploring ~751K Israeli Supreme Court documents. Provides full-text search, corpus statistics with visualizations, on-demand LLM-powered NER analysis, a REST API, and an admin backend.

Built as a deliverable of an Israel Innovation Authority research project (grants #78560, #78561), conducted by the Hebrew University of Jerusalem and Tel Aviv University.

Features

• Full-text search across 733K+ documents using SQLite FTS5
• Faceted filtering by year, document type, legal division, judge, lawyer, party
• Document viewer with proper RTL Hebrew text rendering
• Case view with document timeline and participants (judges, lawyers, parties)
• Statistics dashboard — interactive charts (documents by year, type distribution, top judges/lawyers, technical ratio)
• NER analysis — on-demand LLM-powered Named Entity Recognition with inline highlighting
• REST API with API key authentication and rate limiting
• Admin backend — dashboard, API key management, data import, LLM configuration

Tech Stack

• Backend: Python 3.11+, FastAPI, SQLAlchemy, SQLite + FTS5
• Frontend: React, TypeScript, Tailwind CSS, Recharts
• Package management: uv (Python), npm (frontend)
• LLM integration: LiteLLM (supports Claude, GPT-4, Gemini, Ollama, etc.)
• Deployment: Docker Compose

Prerequisites

• uv (v0.8+)
• Node.js (v20+)
• ~6 GB disk space for the database

Quick Start (Local)

1. Install dependencies

# Python dependencies
uv sync

# Frontend dependencies
cd frontend && npm install && cd ..

2. Configure environment

cp .env.example .env

Edit .env as needed. The defaults work for local development (except LLM features which require an API key).

3. Import the dataset

The dataset is a 1.5 GB Parquet file. If you have it locally at docs/cases_all.parquet:

uv run python backend/import_dataset.py --db data/court.db --source docs/cases_all.parquet

Or download directly from HuggingFace (~5 GB):

uv run python backend/import_dataset.py --db data/court.db

Import takes ~5 minutes and produces a ~5 GB SQLite database with FTS5 indexes.

4. Build the frontend

cd frontend && npm run build && cd ..

5. Start the server

uv run uvicorn backend.app.main:app --host 0.0.0.0 --port 8018

Open http://localhost:8000 in your browser.

Development Mode

For development with hot reload, run backend and frontend separately:

# Terminal 1 — Backend (auto-reloads on Python changes)
uv run uvicorn backend.app.main:app --reload --port 8000

# Terminal 2 — Frontend dev server (hot reload, proxies API to backend)
cd frontend && npm run dev

Then open http://localhost:5173 (the Vite dev server).

Docker

cp .env.example .env
# Edit .env as needed

# Import dataset (one-time)
docker compose run --rm app uv run python backend/import_dataset.py --db /data/court.db

# Start the platform
docker compose up -d

The database is persisted in ./data/ via volume mount.

Configuration

All configuration is via environment variables (or .env file):

Variable	Default	Description
DATABASE_PATH	./data/court.db	Path to SQLite database
ADMIN_USERNAME	admin	Admin login username
ADMIN_PASSWORD	changeme	Admin login password
LLM_MODEL	claude-sonnet-4-20250514	LiteLLM model for NER
LLM_API_KEY	—	API key for the LLM provider
LLM_MAX_TOKENS	4096	Max output tokens for NER
LLM_TEMPERATURE	0.0	LLM temperature
LLM_MAX_INPUT_TOKENS	8000	Max document tokens sent to LLM
NER_RATE_LIMIT_PER_HOUR	10	NER requests per hour per IP
API_RATE_LIMIT_PER_HOUR	100	API requests per hour per key
HOST	0.0.0.0	Server bind host
PORT	8000	Server bind port

API

Interactive API documentation is available at http://localhost:8000/docs (Swagger UI).

Key Endpoints

Method	Endpoint	Description
GET	/api/v1/documents	Search/list documents (FTS5, filterable, paginated)
GET	/api/v1/documents/{hash}	Get document by SHA-256 hash
GET	/api/v1/documents/{hash}/text	Get raw document text
POST	/api/v1/documents/{hash}/ner	Run NER analysis
GET	/api/v1/cases	Search/list cases
GET	/api/v1/cases/{case_desc}	Get case with documents and participants
GET	/api/v1/stats/overview	Corpus-level statistics
GET	/api/v1/stats/by-year	Document counts by year
GET	/api/v1/stats/judges	Top judges by case count
GET	/api/v1/stats/lawyers	Top lawyers by appearance count

Query Parameters

Parameter	Type	Description
q	string	Full-text search query
year_from / year_to	int	Year range filter
type	string	Document type (החלטה or פסק-דין)
division	string	Legal division
judge / lawyer / party	string	Name filter (partial match)
technical	bool	Technical documents only
page / per_page	int	Pagination (default: 1 / 20, max per_page: 100)
sort	string	Sort by: date, year, pages, relevance

Project Structure

website/
├── backend/
│   ├── app/
│   │   ├── main.py              # FastAPI entry point
│   │   ├── config.py            # Settings from env vars
│   │   ├── database.py          # SQLite connection
│   │   ├── models.py            # SQLAlchemy ORM models
│   │   ├── schemas.py           # Pydantic request/response models
│   │   ├── auth.py              # Admin session + API key auth
│   │   ├── routers/
│   │   │   ├── cases.py         # /api/v1/cases
│   │   │   ├── documents.py     # /api/v1/documents
│   │   │   ├── stats.py         # /api/v1/stats/*
│   │   │   ├── ner.py           # /api/v1/documents/{hash}/ner
│   │   │   └── admin.py         # /admin/api/*
│   │   └── services/
│   │       ├── search.py        # FTS5 search logic
│   │       └── ner_service.py   # LiteLLM NER integration
│   └── import_dataset.py        # ETL: Parquet → SQLite
├── frontend/
│   └── src/
│       ├── pages/               # Route-level components
│       ├── components/          # Reusable UI components
│       └── api/                 # API client + types
├── data/                        # SQLite database (created by import)
├── pyproject.toml
├── Dockerfile
└── docker-compose.yml

Data Sources

• Dataset: LevMuchnik/SupremeCourtOfIsrael on HuggingFace
• Legal-HeBERT: github.com/avichaychriqui/Legal-HeBERT

Acknowledgments

Funded by the Israel Innovation Authority (grants #78560, #78561) under the "Kamin" track for applied academic research. Conducted by the Hebrew University of Jerusalem (Prof. Lev Muchnik) and Tel Aviv University (Dr. Inbal Yahav Shenberger).