app.py

from dotenv import load_dotenv

# Load environment variables from .env file
load_dotenv()

from langchain.embeddings import OpenAIEmbeddings  # for creating embeddings
from langchain.vectorstores import Chroma  # for the vectorization part
from langchain.chains import ConversationalRetrievalChain
from langchain.llms import OpenAI  # the LLM model we'll use (CHatGPT)
import gradio as gr

max_sources = 4
DB_DIR = "chroma"

embedding = OpenAIEmbeddings()
vectordb = Chroma(persist_directory=DB_DIR, embedding_function=embedding)
pdf_qa = ConversationalRetrievalChain.from_llm(
    OpenAI(temperature=0.9, model_name="gpt-3.5-turbo"),
    vectordb.as_retriever(),
    return_source_documents=True,
)


def chat_pdf(query, chat_history=""):
    result = pdf_qa({"question": query, "chat_history": chat_history})
    answer = result["answer"]
    source_docs = result["source_documents"]

    print("source_docs", len(source_docs))

    cleaned_docs = []
    for doc in source_docs:
        cleaned_content = doc.page_content
        metadata_info = f"Metadata: {doc.metadata}\n"
        cleaned_docs.append(metadata_info + cleaned_content)

    # Pad the outputs to match the number of output components in the Gradio interface
    padded_outputs = [answer] + cleaned_docs + [""] * (max_sources - len(cleaned_docs))
    return padded_outputs


def create_outputs(num_sources):
    outputs = [gr.outputs.Textbox(label="Answer")]

    for i in range(1, num_sources + 1):
        outputs.append(gr.outputs.Textbox(label=f"Source Document {i}"))

    return outputs


iface = gr.Interface(
    fn=chat_pdf,
    inputs=[gr.inputs.Textbox(label="Query")],
    outputs=create_outputs(max_sources),
    examples=[
        ["Give 2 species of fulgoroidea"],
        ["What colors are found among fulgoroidea?"],
        ["Why are fulgoroidea so cute?"],
    ],
)

iface.launch(debug=True)