This template allows you to retrieve documents using hybrid search, that combines both similarity and full-text search.
It creates an Edge Application and an Edge Function powered by EdgeAI and EdgeSQL database to store and retrieve your data.
git clone https://github.com/PedroMiolaSilva/demo-memedbrew install azionFor other platforms (Windows/Linux) you can view the instructions on how to install the Azion CLI.
Next, add a personal token to authenticate.
azion -t PERSONAL_TOKENAlso, include your token into .env.
Make sure you have Node 20 or higher and yarn installed.
Run the following on the project root folder.
azion link #here select typescript as preset
azion deploy --localAfter a couple minutes, your template will be deployed on Azion's Edge (a domain will be provided at the end of the deployment).
Then, you can test two available routes:
- /query: to retrieve documents based on a query
Search for documents using vector similarity and full-text search.
curl -X POST 'https://your-edge-url/query' \
-H 'Content-Type: application/json' \
-d '{
"query": "your search query",
"filter": "optional_specialty_filter",
"dbName": "optional_database_name" (defaults to memeddatabase)
}'Soma examples of queries are:
To test Full text search capabilities -> query: minox -> query: minoxidil -> query: edemas
To test similarity search capabilities -> query: cabelo caindo -> query: paciente relata dor
You can also specify filters to be used in the "especialidade" column -> filter: Cardiologista -> filter: Ortopedista -> filter: Dermatologista
To enhance the search results, many different strategies can be used on the SQL queries, combining both FTS and vector functionalities. You can access the sqlite documentation to understand how to implement these: sqlite docs.
The files used on the demonstration are AI generated. They are in migrations/files/medication.json.
Note: The dbName parameter is optional in both endpoints. If not provided, it defaults to 'memeddatabase'.