This repository contains the code for a simple implementation of a vector database. It can absorb pdf files, which are split into groups of sentences and then vectorized. The vectors are stored in a Weaviate database, which can be queried for similar sentences.
The database then can be queried for similar sentences to a given sentence or a question. The similarity is calculated by the cosine similarity of the vectors. The results are then displayed in a simple web interface. In addition, the potential answer is highlighted in the original text.
Models used are:
sentence-transformers/multi-qa-mpnet-base-dot-v1- This model is used to calculate the vectors of the sentences and questions.
bert-large-uncased-whole-word-masking-finetuned-squad- This model is used to find and highlight the answer to a question in a given text.
Set up a virtual environment with
python3 -m venv .venv
Activate the virtual environment with
source .venv/bin/activate
Install Python requirements with
pip3 install -r requirements.txt
Start the Weaviate server at localhost:8080 with
docker-compose up -d --build
Run the setup of the Weaviate database with
python3 setup.py
Start the virtual environment with (if not already active)
source .venv/bin/activate
Start the Streamlit app with
streamlit run app.py
Start the Docker container with (if not already running)
docker-compose up -d --build
Stop the Docker container with
docker-compose down
Deactivate the virtual environment with
deactivate
If you want to access the app from outside your local network, you can use ngrok. Follow the installation instructions on their website.
After that, make sure Streamlit is running and you can start the tunnel with
ngrok http http://localhost:8501