Use the GPT-3.5 Turbo api to build a chatGPT chatbot for multiple Large PDF files.
Tech stack used includes LangChain, Pinecone, Typescript, OpenAI, and Next.js. LangChain is a framework that makes it easier to build scalable AI/LLM apps and chatbots. Pinecone is a vectorstore for storing embeddings and your PDF in text to later retrieve similar docs.
Prelude: Please make sure you have already downloaded node on your system and the version is 18 or greater.
- Install packages
First run npm install yarn -g to install yarn globally (if you haven't already).
Then run:
yarn install
After installation, you should now see a node_modules folder.
- Set up your
.envfile
- Copy
.env.exampleinto.envYour.envfile should look like this:
OPENAI_API_KEY=
PINECONE_API_KEY=
PINECONE_ENVIRONMENT=
PINECONE_INDEX_NAME=
- Visit openai to retrieve API keys and insert into your
.envfile. - Visit pinecone to create and retrieve your API keys, and also retrieve your environment and index name from the dashboard.
-
In the
configfolder, replace thePINECONE_NAME_SPACEwith anamespacewhere you'd like to store your embeddings on Pinecone when you runnpm run ingest. This namespace will later be used for queries and retrieval. -
In
utils/makechain.tschain change theQA_PROMPTfor your own usecase.
This repo can load multiple PDF files
-
Inside
docsfolder, add your pdf files or folders that contain pdf files. -
Run the script
npm run ingestto 'ingest' and embed your docs. If you run into errors troubleshoot below. -
Check Pinecone dashboard to verify your namespace and vectors have been added.
Once you've verified that the embeddings and content have been successfully added to your Pinecone, you can run the app npm run dev to launch the local dev environment, and then type a question in the chat interface.
The code has been written referring the following YouTube tutorial: https://www.youtube.com/watch?v=ih9PBGVVOO4 and referred the GitHub repo: https://github.com/mayooear/gpt4-pdf-chatbot-langchain for understanding and production of the whole chatbot. Thanks to him!