app.py

import gradio as gr
from huggingface_hub import InferenceClient
from typing import List, Tuple
import fitz  # PyMuPDF
from sentence_transformers import SentenceTransformer, util
import numpy as np
import faiss

client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")

# Placeholder for the app's state
class MyApp:
    def __init__(self) -> None:
        self.documents = []
        self.embeddings = None
        self.index = None
        self.load_pdf("THEDIA1.pdf")
        self.build_vector_db()

    def load_pdf(self, file_path: str) -> None:
        """Extracts text from a PDF file and stores it in the app's documents."""
        doc = fitz.open(file_path)
        self.documents = []
        for page_num in range(len(doc)):
            page = doc[page_num]
            text = page.get_text()
            self.documents.append({"page": page_num + 1, "content": text})
        print("PDF processed successfully!")

    def build_vector_db(self) -> None:
        """Builds a vector database using the content of the PDF."""
        model = SentenceTransformer('all-MiniLM-L6-v2')
        self.embeddings = model.encode([doc["content"] for doc in self.documents])
        self.index = faiss.IndexFlatL2(self.embeddings.shape[1])
        self.index.add(np.array(self.embeddings))
        print("Vector database built successfully!")

    def search_documents(self, query: str, k: int = 3) -> List[str]:
        """Searches for relevant documents using vector similarity."""
        model = SentenceTransformer('all-MiniLM-L6-v2')
        query_embedding = model.encode([query])
        D, I = self.index.search(np.array(query_embedding), k)
        results = [self.documents[i]["content"] for i in I[0]]
        return results if results else ["No relevant documents found."]

app = MyApp()

def preprocess_input(user_input: str) -> str:
    """Preprocesses user input to enhance it for better context."""
    if "therapy" in user_input.lower():
        return "I am looking for guidance on therapy. Can you help me with some exercises or techniques to manage my stress and emotions?"
    # Add more rules as needed
    return user_input

def preprocess_response(response: str) -> str:
    """Preprocesses the response to make it more polished."""
    response = response.strip()
    response = response.replace("\n\n", "\n")
    response = response.replace(" ,", ",")
    response = response.replace(" .", ".")
    response = " ".join(response.split())
    return response

def shorten_response(response: str) -> str:
    """Uses the Zephyr model to shorten and refine the response."""
    messages = [{"role": "system", "content": "Shorten and refine this response."}, {"role": "user", "content": response}]
    result = client.chat_completion(messages, max_tokens=256, temperature=0.5, top_p=0.9)
    return result.choices[0].message['content'].strip()

def respond(message: str, history: List[Tuple[str, str]]):
    system_message = "You are a concisely speaking empathetic Dialectical Behaviour Therapist assistant. You politely guide users through DBT exercises based on the given DBT book. You must say one thing at a time and ask follow-up questions to continue the chat."
    messages = [{"role": "system", "content": 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]})

    # Preprocess user input
    preprocessed_message = preprocess_input(message)
    messages.append({"role": "user", "content": preprocessed_message})

    # RAG - Retrieve relevant documents
    retrieved_docs = app.search_documents(preprocessed_message)
    context = "\n".join(retrieved_docs)
    if context.strip():
        messages.append({"role": "system", "content": "Relevant documents: " + context})

    response = client.chat_completion(messages, max_tokens=1024, temperature=0.7, top_p=0.9)
    response_content = "".join([choice.message['content'] for choice in response.choices if 'content' in choice.message])
    
    polished_response = preprocess_response(response_content)
    shortened_response = shorten_response(polished_response)

    history.append((message, shortened_response))
    return history, ""

with gr.Blocks() as demo:
    gr.Markdown("# 🧘‍♀️ **Dialectical Behaviour Therapy**")
    gr.Markdown(
        "‼️Disclaimer: This chatbot is based on a DBT exercise book that is publicly available. "
        "We are not medical practitioners, and the use of this chatbot is at your own responsibility."
    )

    chatbot = gr.Chatbot()

    with gr.Row():
        txt_input = gr.Textbox(
            show_label=False,
            placeholder="Type your message here...",
            lines=1
        )
        submit_btn = gr.Button("Submit", scale=1)
        refresh_btn = gr.Button("Refresh Chat", scale=1, variant="secondary")

    example_questions = [
        ["I feel overwhelmed with work."],
        ["Can you guide me through a quick meditation?"],
        ["How do I stop worrying about things I can't control?"],
        ["What are some DBT skills for managing anxiety?"],
        ["Can you explain mindfulness in DBT?"],
        ["What is radical acceptance?"],
        ["How can I practice distress tolerance?"],
        ["What are some techniques to handle distressing situations?"],
        ["How does DBT help with emotional regulation?"],
        ["Can you give me an example of an interpersonal effectiveness skill?"]
    ]

    gr.Examples(examples=example_questions, inputs=[txt_input])

    submit_btn.click(respond, [txt_input, chatbot], [chatbot, txt_input])
    refresh_btn.click(lambda: [], None, chatbot)

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