Quickstart

This demo is packaged with a sample data set that was used for a custom "find a doctor" demo that was originally built to augment a customer workshop. This code can be modified to support ANY use case. The demo showcases the following features:

1. Suggesters - AKA autocomplete
2. Full Text Search - "profile", "specialty", "title", "gender", "address", "phone"
3. Semantic Search - "profile" vectorized via E5
4. Elastic Reranker or RRF - You can toggle between the 2 options by changing CROSS_ENCODER_ENABLED in app.py

Setup

Create a .env file that has the following properties:

ES_URL=YOUR_ES_URL
ES_API_KEY=YOUR_API_KEY

Open a terminal and run the following commands

Install dependencies

pip install poetry
poetry install

Setup Elasticsearch indexes

python elastic_index_manager.py setup

Run web application

poetry run uvicorn app:app --reload 

Open a browser to http://localhost:8000

Type the following text to trigger multiple suggestions:

imm

Type the following text to trigger a single suggestion with fuzzy match:

hart

Change from elastic reranker to RRF by editing app.py. Change the value of CROSS_ENCODER_ENABLED from True to `False'. You can also change the reranker settings to alter the behavior or RRF or cross encoder reranker.

Teardown

CTRL+C the terminal shell that is running the fastapi application.

Delete the indexes in Elasticsearch

python elastic_index_manager.py teardown

Customize with different data

It's very hard to create an end to end demo with swappable data but you can alter the code to work with a different use case with a few tweaks. This demo features a super simplistic web front end. It is just a single index.html page with embedded CSS styles and javascript for making calls to /search and /autocomplete. Customizing this demo involves the following:

1. Modify ./suggest_index.jsonl with the suggestions that you want in suggest index.
2. Modify ./search_index.jsonl with the data that you want in the search index.

If you just want to modify suggestions, you can hand edit ./suggest_index.jsonl and rerun python elastic_index_manager.py reload. There is no need to restart the application. Your new suggestions should display.

If you change the field names/types in search_index.jsonl

3. You will need to update the /search function in ./app.py
4. You will need to update recordDiv.innerHTML in ./static/index.html