Langchain Explorations

https://twitter.com/olshansky/status/1672827218563850240

• Getting started
  • Prepare Your Secrets
  • Option 1: Vitalik Articles (default)
  • Option 2: Your data (custom)
• References & Resources
  • Startup Stack to easily productionize
• Future Work / TODO / Improvements

Getting started

Prepare Your Secrets

1. Get your OpenAI key from here; make sure you're logged in
2. Get your PinconeAI key from here; make sure you're logged in
3. Create an index called my-index or vitalik-index as showing here depending on which option you choose below

cp .streamlit/secrets.toml.template .streamlit/secrets.toml

Update the env vars appropriately.

Option 1: Vitalik Articles (default)

1. Start the virtual environment and install the requirements:

$(make env_source)
make pip_install

2. Download Vitalik's articles

make download_vitalik_articles

3. Create the embeddings and upload them to Pinecone

make prepare_db

4. Run it locally and ask a question:

make streamlit_app

Option 2: Your data (custom)

1. Start the virtual environment and install the requirements:

$(make env_source)
make pip_install

2. Get your pdfs and put them in ./data/my_data

    mv or cp or w/e floats your boat into `./data/my_data`

3.Prepare the database with your pdfs

# DATADIR and INDEX_NAME are customizable but you can start with the defaults if you don't know what you're doing yet
make prepare_db DATADIR=data/my_data INDEX_NAME=my-index

4. Open up query_db.py and update prompt_template appropriately

5. Run it locally and ask a question:

make streamlit_app

References & Resources

1. Initial reference for this repo: https://blog.bytebytego.com/p/how-to-build-a-smart-chatbot-in-10
2. Langchain docs: https://python.langchain.com/
3. A full book I haven't read yet: https://www.pinecone.io/learn/langchain-intro/
4. OpenAI API: https://beta.openai.com/docs/introduction

Startup Stack to easily productionize

• Easily host infra: fly.io or render.com
• Easily deploy WIP app in python: https://streamlit.io/generative-ai
• Building a backend API w/ nothing: https://github.com/fern-api/fern

Future Work / TODO / Improvements

• Notion - Add support for pulling data from notion
• Twitter - Use user tweet accounts to do this as well
• OpenAI - Use the OpenAI Chat Completion API directly
• Pinecone - Need to be able to update existing index