LegalAI

A Retrieval-Augmented Generation (RAG) system for Canadian case law analysis. Ask natural language questions about Supreme Court and CanLII decisions and receive grounded, citation-backed answers.

Features

• Semantic Search: Vector-based retrieval using OpenAI embeddings (1536D)
• Intelligent Case Detection: Automatically detects and prioritizes specific cases mentioned in queries
• CanLII Citation Formatting: Proper legal citation format for all retrieved cases
• Token Budget Management: Smart context fitting (max 4000 tokens)
• Source Attribution: All answers include case citations and relevance scores
• Interactive CLI: Ask questions in natural language with real-time retrieval

Technology Stack:

• Vector Database: ChromaDB with persistent storage
• Embeddings: OpenAI text-embedding-3-small (1536 dimensions)
• LLM: OpenAI gpt-4o-mini for answer generation
• Chunking: tiktoken with cl100k_base encoding
• Retrieval: Euclidean distance with HNSW indexing

Installation

Prerequisites

• Python 3.10+
• OpenAI API key

Setup

1. Clone the repository:

git clone <repository-url>
cd LegalAI

2. Create virtual environment:

python3 -m venv venv
source venv/bin/activate  # On macOS/Linux
# venv\Scripts\activate   # On Windows

3. Install dependencies:

pip install -r requirements.txt

4. Configure environment: Create a .env file in the project root:

OPENAI_API_KEY=your_api_key_here

Usage

Quick Start

Run the interactive Q&A system:

python app/agent/rag_pipeline.py

Example queries:

• "What is the standard for grossly disproportionate sentences?"
• "What was R. v. Vrbanic about?"
• "What is the most recent Supreme Court judgment in 2025?"
• "Explain the test for unreasonable search and seizure"

Features:

• Case name detection: "R. v. [Name]" → retrieves all chunks from that case
• Citation detection: "2025 SCC 38" → prioritizes that specific case
• General queries: Semantic search across all cases

Example Output

Your question: What is the standard for grossly disproportionate sentences?

Retrieving relevant case law for: "What is the standard for grossly disproportionate sentences?"
Sources: 2023 SCC 26 (CanLII), 2024 SCC 38 (CanLII), 2025 SCC 41 (CanLII)

ANSWER:

The standard for 'grossly disproportionate' sentences is described as being "so excessive 
as to outrage standards of decency" and "abhorrent or intolerable" to society. The standard 
is intended to be demanding and will be attained only rarely. To establish that a sentence 
is grossly disproportionate, it must be shown that the minimum sentence goes beyond what 
is necessary to achieve Parliament's objectives relevant to the offence.

Sources (3 cases):
  - 2023 SCC 26 (CanLII)
  - 2024 SCC 38 (CanLII)
  - 2025 SCC 41 (CanLII)
Context: 4 chunks, 3847 tokens
Model: gpt-4o-mini-2024-07-18 | Tokens: 4207

Known Limitations

Data Completeness

• Oral judgments: Some recent cases (e.g., R. v. Vrbanic, 2025 CanLII 125773) contain only oral summaries with "Reasons to follow"
• Solution: Manually download full written reasons from CanLII when published and re-run ingestion pipeline

Context Window

• Limited to 4000 tokens (~3-5 pages of case law)
• Very long cases may be truncated
• Complex multi-case questions may lack full context

Retrieval Accuracy

• Semantic search may occasionally miss relevant chunks
• Requires well-formed natural language queries
• Legal terminology recognition depends on training data

Not Legal Advice

• This system summarizes case law contextually
• Does not provide legal advice or opinions
• Users must verify all information independently
• Consult a qualified lawyer for legal decisions

Legal & Compliance

CanLII Terms of Use

• Individual case downloads for research: Permitted
• Academic and research use with attribution: Permitted
• Bulk downloading or systematic scraping: Prohibited
• Commercial use without permission: Prohibited

Attribution

All cases sourced from CanLII.org - Canadian Legal Information Institute.

Disclaimer: This tool is for research and educational purposes only. It does not constitute legal advice. Always consult a qualified legal professional for legal matters.

---

Made with ❤️ by Mithish :)