app.py

import gradio as gr
from huggingface_hub import InferenceClient

# RAG imports
import os
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
from langchain.schema import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter

"""
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
"""
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")

# We'll load the existing FAISS index at the start
INDEX_FOLDER = "faiss_index"
_vectorstore = None

def load_vectorstore():
    """Loads FAISS index from local folder."""
    global _vectorstore
    if _vectorstore is None:
        embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
        _vectorstore = FAISS.load_local(INDEX_FOLDER, embeddings, allow_dangerous_deserialization=True)
    return _vectorstore

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    """
    Called on each user message. We'll do a retrieval step (RAG)
    to get relevant context, then feed it into the system message
    before calling the InferenceClient.
    """
    # 1. Retrieve top documents from FAISS
    vectorstore = load_vectorstore()
    top_docs = vectorstore.similarity_search(message, k=3)

    # Build context string from the docs
    context_texts = []
    for doc in top_docs:
        context_texts.append(doc.page_content)
    KnowledgeBase = "\n".join(context_texts)

    # 2. Augment the original system message with retrieved context
    augmented_system_message = system_message + "\n\n" + f"Relevant context:\n{KnowledgeBase}"

    # 3. Convert (history) into messages
    messages = [{"role": "system", "content": augmented_system_message }]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

    # Finally, add the new user message
    messages.append({"role": "user", "content": message})

    # 4. Stream from the InferenceClient
    response = ""

    for message in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message.choices[0].delta.content

        response += token
        yield response


"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly, knowledgeable assistant acting as Prakash Naikade." 
                "You have access to a rich set of documents and references collectively called KnowledgeBase, which you should call and treat as your current knowledge base. "
                "Always use the facts, details, and stories from KnowledgeBase to ground your answers. "
                "If a question goes beyond what KnowledgeBase covers, politely explain that you don’t have enough information to answer. "
                "Remain friendly, empathetic, and helpful, providing clear, concise, and context-driven responses. "
                "Stay consistent with any personal or professional details found in KnowledgeBase. "
                "If KnowledgeBase lacks any relevant detail, avoid making up new information—be honest about the gap. "
                "Your goal is to accurately represent Prakash Naikade: his background, expertise, and experiences, using only the data from KnowledgeBase to support your answers.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)


if __name__ == "__main__":
    demo.launch()