SemanticSearch

Minimal RAG (Retrieval Augmented Generation) website with Pinecone, FastAPI, NextJS, and MongoDB.

SemanticSearchDemo.mov

Install

You might want to use pyenv to create a virtualenv in which to install Python dependencies:

$ pyenv virtualenv 3.10.1 myenvname
$ pyenv activate myenvname
(myenvname) $ pip install -r requirements.txt

Assuming you have Node/npm (perhaps with nvm), simply clone this repo, and then from the ./ui folder (with package.json in it) install dependencies:

$ cd ./ui
$ npm install 

Run

Web App

Start the server (port 8000):

$ cd ./backend
$ uvicorn backend.main:app --reload

Now you're ready to run the server locally (port 3000) from within ./ui:

$ cd ./ui
$ npm run dev

And just head over to http://localhost:3000 to demo!

Test APIs with Curl

Ask a question

$ curl -X POST http://localhost:8000/query \
    -H 'Content-Type: application/json' \
    -d '{"query": "MY QUESTION HERE"}'

(didn't work) Upload a file https://superuser.com/questions/1054742/how-to-post-file-contents-using-curl

$ curl -X POST http://localhost:8000/upload \
    --data-binary @data/essays/ace.txt 

Take 2 https://medium.com/@petehouston/upload-files-with-curl-93064dcccc76

$ curl http://localhost:8000/upload \
    -F "files=@data/essays/ace.txt"

Send multiple files at once:

$ curl -X POST http://localhost:8000/upload \
    -H "Content-Type: multipart/form-data" \
    -F "files=@data/essays/ace.txt" \
    -F "files=@data/essays/top.txt"

Load everything in ./data/essays (WARNING: this will take ~10 minutes)

$ curl -X POST http://localhost:8000/reupload_all

Testing

Use pytest with to run tests in ./backend/test_main.py:

$ cd ./backend
$ python -m pytest -v

For a specific test:

$ python -m pytest -v test_main.py::test_step0 

For some reason pytest has weird behavior with relative/absolute imports, which prevents you from running pytest directly, but the above should be fine.

Misc

Pinecone

https://docs.pinecone.io/docs/quickstart

$ curl -i https://controller.us-west1-gcp.pinecone.io/actions/whoami -H 'Api-Key: MY_KEY'

MongoDB

CLI (after adding your IP address and creating a user):

$ brew install mongosh
$ mongosh "mongodb+srv://cluster0.hlqhjo4.mongodb.net/myFirstDatabase" --apiVersion 1 --username nsbradford

Future Work

This is built with FastAPI, NextJS, MongoDB, and Pinecone.

Next:

• hyperlink to raw document
• oneclick "explore more", combine found vector
• threshold warning color result likely quality
• NER