AI Swiss Legal Assistant 🇨🇭 👩‍⚖️ 🤖

This is a simple conversational-ui RAG (retrieval augmented generation) based on the Swiss Code of Obligations and additional Swiss jurisprudence trilingual dataset (in English, German and French).

It was created a starting point of the Ginetta Challenge at the women++ Hack'n'Lead hackathon November 2023

Conversational AI model used: Mistral

Source of data used to re-trained the model: https://entscheidsuche.ch/docs (PDF files of Swiss jurisprudence).

Steps and associated files

• Text extraction from pdf and write in into a json file → docs/download_pdf.py & docs/extract_texts_from_pdf_and_save_it_to_json.py
• Preprocessing → docs/data-cleaning-fromjsontojson.py
• Embeddings using https://huggingface.co/Xenova/paraphrase-multilingual-mpnet-base-v2 → create_embeddings/create_embeddings.js
• Prompt engineering aimed at better legal assistance → app/api/chat/route.ts

To launch the chatbot app locally

1. 🦙 Download Ollama and install it locally

2. Run the Mistral model using: ollama run mistral (Requires 4.1GB and 8GB of RAM)

3. Open http://localhost:11434 to check if Ollama is running

4. docker pull qdrant/qdrant

5. docker run -p 6333:6333 qdrant/qdrant

6. Open the Qdrant dashboard console http://localhost:6333/dashboard#/console

7. Create a new collection running this:

PUT collections/swiss-or { "vectors": { "size": 384, "distance": "Cosine" } }

8. Download our new custom snapshots: swiss-civil-code-de-paraphrase-multilingual-mpnet-base-v2.zip , swiss-code-of-obligations-en-paraphrase-multilingual-mpnet-base-v2.zip

9. Unzip the file using the terminal

10. Load our snapshot to Qdrant using the following code:

curl -X POST 'http://localhost:6333/collections/swiss_law/snapshots/upload' -H 'Content-Type:multipart/form-data' -F 'snapshot=@swiss-civil-code-de-paraphrase-multilingual-mpnet-base-v2.snapshot'

curl -X POST 'http://localhost:6333/collections/swiss_law/snapshots/upload' -H 'Content-Type:multipart/form-data' -F 'snapshot=@swiss-code-of-obligations-en-paraphrase-multilingual-mpnet-base-v2.snapshot'

11. Copy the file .env.local.example and rename it to .env. Verify that all environment variables are correct.
12. yarn install to install the required dependencies
13. yarn dev to launch the development server

Finally -> Go to http://localhost:3000 and try out the app