PFF lets you upload pdf files and ask an LLM (currently ChatGPT) questions in a chat interface about the pdf's materials. It's live, you can make an account and started using it.

Run Locally

clone the repo

git clone https://github.com/hfkhrni/pff.git

cd into the repo's folder and run pnpm install

run a deveopment server

pnpm run dev

rename the .env.example file to .env after filling it out

pinecone requires the index name as a string and you'll need to put your index name in src/app/api/message/route.ts and src/app/api/uploadthing/core.ts where the index variable is declared

  const pineconeIndex = pinecone.Index('index name here');

Other Filetypes

you can use this for files other than pdfs, there are 3 parts to this:

1. change the filetype allowed in uploadthing's endpoint in src/app/api/uploadthing/core.ts

export const ourFileRouter = {
  pdfUploader: f({ "file-type-here": { maxFileSize: '<number>MB' } })
}

2. pdfs are rendered by the <PDFRenderer /> component inside <ChatWrapper />, you'll need to create a component that views your file next to the chat if you need that.
3. configure the prompt in src/app/api/message/route.ts to suite the filetype you're querying

Replace ChatGPT

there's only one endpoint which calls ChatGPT's API with the embeddings and gets back a streaming response. this endpoint is src/app/api/message/route.ts.

you will need to get embeddings from the model you're using in replacement of:

  const embeddings = new OpenAIEmbeddings({
    openAIApiKey: process.env.OPENAI_API_KEY,
  });

and where you call the api:

  const openai = new OpenAI({
      apiKey: process.env.OPENAI_API_KEY,
    });
  const response = await openai.chat.completions.create({
    model: 'gpt-3.5-turbo',
    temperature: 0,
    stream: true,
    messages: []
})