Neural search demo
With Qdrant + BERT + FastAPI
This repository contains a code for Neural Search for startups demo.
The demo is based on the vector search engine Qdrant.
Requirements
Install python requirements:
pip install poetry
poetry install
You will also need Docker and docker-compose
Quick Start
To launch this demo locally you will need to prepare data first.
The source of the original data is https://www.startups-list.com/
Code for initial data preparation could be found in Colab Notebook.
After evaluating Colab you should get startup records in file ./data/startups.json and encoded vectors in file ./data/startup_vectors.npy
Generate full-text index in ./data/startups.sqlite3
# Init full-text index
python -m qdrant_demo.init_text_search_index
To launch service locally, use
docker-compose -f docker-compose-local.yaml up
After service is started you can upload initial data to the search engine.
# Init neural index
python -m qdrant_demo.init_vector_search_index
After a successful upload, neural search API will be available at http://localhost:8000/docs