Ethio Legal Assistant

AI-powered legal question assistant for Ethiopian laws using Large Language Models (LLMs) and Retrieval Augmented Generation (RAG).

Project Vision

Legal information in Ethiopia is often difficult to access, scattered across PDFs, and hard to search.
This project builds a local AI assistant that allows users to ask legal questions in natural language and receive answers grounded in Ethiopian legal documents.

Example questions:

• What are employee rights under Ethiopian labor law?
• What is the legal process for land disputes?
• What penalties exist for contract violations?

---

Features

• Natural language legal Q&A
• Retrieval from Ethiopian legal documents
• Local LLM inference (CPU friendly)
• Source citations from legal texts
• Modular architecture

Planned:

• Amharic language support
• Voice interface
• Legal document summarization
• Case law search

---

Architecture

User Question
↓
Embedding Model
↓
Vector Database Search
↓
Relevant Legal Documents
↓
LLM Generates Answer
↓
Response with Sources

Main components:

1. Document processing pipeline
2. Vector database
3. Embedding model
4. Local LLM
5. API / Interface

---

Technology Stack

• Python
• LangChain
• FAISS Vector Database
• SentenceTransformers
• Local LLM (Mistral / Llama / Phi)
• FastAPI
• Streamlit

Optional tools:

• Ollama
• HuggingFace Transformers
• ChromaDB

---

Project Structure

ethio-legal-assistant
│
├── data
│   ├── raw_legal_docs
│   └── processed_docs
│
├── embeddings
│   └── build_vector_db.py
│
├── rag
│   ├── retriever.py
│   ├── prompt_template.py
│   └── pipeline.py
│
├── models
│   └── load_llm.py
│
├── api
│   └── main.py
│
├── app
│   └── streamlit_app.py
│
├── notebooks
│   └── experiments.ipynb
│
├── requirements.txt
└── README.md

---

Step 1 --- Collect Ethiopian Legal Documents

Gather public legal documents such as:

• Ethiopian Constitution
• Labor Proclamation
• Civil Code
• Criminal Code
• Land laws

Save them in:

data/raw_legal_docs/

---

Step 2 --- Install Environment

Create environment:

python -m venv venv
source venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Example dependencies:

• langchain
• sentence-transformers
• faiss-cpu
• transformers
• accelerate
• pypdf
• fastapi
• uvicorn
• streamlit

---

Step 3 --- Process Legal Documents

Extract text and split into chunks.

python data/process_documents.py

This step performs:

• PDF extraction
• text cleaning
• chunking legal documents

---

Step 4 --- Create Embeddings

Generate vector embeddings.

python embeddings/build_vector_db.py

Result:

vector_store/

---

Step 5 --- Load Local LLM

Recommended models:

• Mistral 7B
• Llama 3
• Phi-2

Example loader:

models/load_llm.py

---

Step 6 --- Build RAG Pipeline

Pipeline steps:

1. Question embedding
2. Vector similarity search
3. Retrieve relevant documents
4. Generate answer using LLM

Main file:

rag/pipeline.py

---

Step 7 --- Run API

uvicorn api.main:app --reload

Example endpoint:

POST /ask

Request example:

{
 "question": "What rights do workers have under Ethiopian labor law?"
}

---

Step 8 --- Launch Interface

streamlit run app/streamlit_app.py

Workflow:

User asks question
System retrieves legal documents
LLM generates response
Sources are shown

---

Example Interaction

Question:

What is the probation period allowed under Ethiopian labor law?

Response:

Explanation of law
Relevant legal clause
Source citation

---

Evaluation Strategy

• Test with curated legal questions
• Measure hallucination rate
• Improve prompts and retrieval
• Verify answers with citations

---

Future Improvements

• Amharic LLM support
• Telegram / WhatsApp chatbot
• Ethiopian legal dataset expansion
• Court case retrieval

---

Ethical Notice

This project provides informational assistance only and does not replace professional legal advice.

Users should consult licensed legal professionals for official legal decisions.

---

Author

Asibeh Tenager
AI Educator & Developer

Focus:

• Machine Learning
• LLM Systems
• AI for African solutions