app.py

import os
import requests
import fitz  # PyMuPDF for PDF reading
import faiss
import numpy as np
import gradio as gr
from sentence_transformers import SentenceTransformer
from huggingface_hub import InferenceClient

# 🔹 Define PDF Directory and Chunk Size
PDF_DIR = "./pdfs"
CHUNK_SIZE = 2500  # Larger chunks for better context

# 🔹 Ensure Directory Exists
os.makedirs(PDF_DIR, exist_ok=True)

# 🔹 Direct URLs for PDF Downloads (Colorado Policy Documents)
PDF_FILES = {
    "SNAP 10 CCR 2506-1.pdf": "https://huggingface.co/spaces/tstone87/ccr-colorado/resolve/main/SNAP%2010%20CCR%202506-1%20.pdf?download=true",
    "Med 10 CCR 2505-10 8.100.pdf": "https://huggingface.co/spaces/tstone87/ccr-colorado/resolve/main/Med%2010%20CCR%202505-10%208.100.pdf?download=true",
}

# 🔹 Function to Download PDFs

def download_pdfs():
    for filename, url in PDF_FILES.items():
        pdf_path = os.path.join(PDF_DIR, filename)
        if not os.path.exists(pdf_path):
            print(f"📥 Downloading {filename}...")
            try:
                response = requests.get(url, stream=True)
                response.raise_for_status()
                with open(pdf_path, "wb") as f:
                    for chunk in response.iter_content(chunk_size=8192):
                        f.write(chunk)
                print(f"✅ Downloaded {filename}")
            except Exception as e:
                print(f"❌ Error downloading {filename}: {e}")

# 🔹 Function to Extract Text from PDFs

def extract_text_from_pdfs():
    all_text = ""
    for pdf_file in os.listdir(PDF_DIR):
        if pdf_file.endswith(".pdf"):
            pdf_path = os.path.join(PDF_DIR, pdf_file)
            doc = fitz.open(pdf_path)
            for page in doc:
                all_text += page.get_text("text") + "\n"
    return all_text

# 🔹 Initialize FAISS Index

def initialize_faiss():
    download_pdfs()
    text_data = extract_text_from_pdfs()
    if not text_data:
        raise ValueError("❌ No text extracted from PDFs!")
    
    chunks = [text_data[i:i+CHUNK_SIZE] for i in range(0, len(text_data), CHUNK_SIZE)]
    model = SentenceTransformer("multi-qa-mpnet-base-dot-v1")
    embeddings = np.array([model.encode(chunk) for chunk in chunks])
    index = faiss.IndexFlatL2(embeddings.shape[1])
    index.add(embeddings)
    print("✅ FAISS index initialized.")
    return index, chunks

# 🔹 Initialize FAISS on Startup
index, chunks = initialize_faiss()

# 🔹 Function to Search FAISS

def search_policy(query, top_k=3):
    query_embedding = SentenceTransformer("multi-qa-mpnet-base-dot-v1").encode(query).reshape(1, -1)
    distances, indices = index.search(query_embedding, top_k)
    return "\n\n".join([chunks[i] for i in indices[0] if i < len(chunks)])

# 🔹 Hugging Face LLM Client
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")

# 🔹 Function to Handle Chat Responses

def respond(message, history):
    messages = [{"role": "system", "content": "You are a chatbot specializing in Colorado public assistance programs."}]
    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})
    
    policy_context = search_policy(message)
    if policy_context:
        messages.append({"role": "assistant", "content": f"📄 **Colorado Policy Info:**\n\n{policy_context}"})
    
    messages.append({"role": "user", "content": message})
    response = ""
    for message in client.chat_completion(messages, max_tokens=512, stream=True, temperature=0.7, top_p=0.95):
        token = message.choices[0].delta.content
        response += token
        yield response

# 🔹 Gradio Chat Interface (Colorado-Themed)
demo = gr.ChatInterface(
    respond,
    textbox=gr.Textbox(placeholder="Ask about Colorado public assistance programs...", interactive=True, show_label=False),
    submit_btn=gr.Button("Send"),
    chatbot=gr.Chatbot(),
)

demo.launch()