Overview:

Author:
Kyle Brandt
GitHub: knb47

Technologies used:
LangChain, Gmail API + Oauth, Pinecone, OpenAI Embeddings + ChatGPT, Streamlit

Have you ever vaguely remembered an email you wanted to retrieve from your inbox but just can't find it? And even after searching for it, you find yourself manually sifting through dozens of search results with no success?

This application can retrieve your emails with only vague details or descriptions. In essence, you no longer need to search for emails using a matching word or a specific date.

When the application finds the email, it will return the exact email subject and date of the email you're looking for, along with some information about the email.

How does this work? This tool stores all of your Gmail emails in Pinecone, a vector database, with the assistance of LangChain + OpenAI embeddings. Upon execution of a search, Pinecone DB will return multiple documents that closely match the email you're looking for using semantic search.

LangChain allows us to "chain" different application components together, such as Pinecone and OpenAI, in a sequential manner. What's also beneficial about LangChain is that it allows us to parse emails into smaller chunks of data to store them in the database efficiently.

Starting interface:

Example of semantic search (using meaning rather than matching words to search):

Getting started:

1. Create a Google developer account, Pinecone account, and OpenAI account.

2. Add the following credentials to a .env file as environment variables:

GOOGLE_APPLICATION_CREDENTIALS = your Google developer account key
OPENAI_API_KEY = key
PINECONE_API_KEY = key
PINECONE_API_ENV = key
PINECONE_INSTANCE_NAME = instance name

For this, you must create a Pinecone DB instance.

How to create a Pinecone instance (Dimensions MUST be 1536 / Metric MUST be Cosine):

Grab the PINECONE_API_ENV from "Environment" as seen below:

3. Using your Google developer account, assign your desired Gmail account as an eligible user for this application in beta mode. If you don't do this, the application won't be able to read your emails since this is a highly-privileged task. I recommend you create a new Gmail account for this application and use mock data to ensure you don't leak sensitive email information to Pinecone or OpenAI.

4. Create a credentials.json file like this below to allow Google Oauth/Gmail API to work.

{ "installed": { "client_id": "key.apps.googleusercontent.com", "project_id": "*project name", "auth_uri": "https://accounts.google.com/o/oauth2/auth", "token_uri": "https://oauth2.googleapis.com/token", "auth_provider_x509_cert_url": "https://www.googleapis.com/oauth2/v1/certs", "client_secret": "client secret", "redirect_uris": [ "http://localhost" ] } }

Running the Application:

1. Run the app by running "streamlit run main.py" in the command line while in the application folder.

2. Click "Authenticate and Extract" to log into your email account. Once authenticated, the app will then begin reading ALL of your emails into the Pinecone Database. This operation may take some time if you have have many emails.

3. Type in your search query into the search bar and hit enter. The app should retrieve the email that best matches your query and provide any relevant information about that email. Remember, you don't need to search for exact strings! Descriptions and details work also, making this a powerful application.

Enjoy!

Inspiration:
Greg Kamradt
LangChain101: Question A 300 Page Book (w/ OpenAI + Pinecone)
https://www.youtube.com/watch?v=h0DHDp1FbmQ&t=563s