app.py

import gradio as gr
import tempfile
from pdfminer.high_level import extract_text
from sentence_transformers import SentenceTransformer
import faiss
from ctransformers import AutoModelForCausalLM

# Embeddings Model
embedder = SentenceTransformer("all-MiniLM-L6-v2")

# FREE, non-gated GGUF model
llm = AutoModelForCausalLM.from_pretrained(
    "TheBloke/OpenHermes-2.5-Mistral-7B-GGUF",
    model_file="openhermes-2.5-mistral-7b.Q4_K_M.gguf",
    model_type="mistral",
    gpu_layers=0
)

doc_chunks, index = [], None

def extract_text_from_pdf(pdf_path):
    return extract_text(pdf_path)

def chunk_text(text, size=500, overlap=50):
    words = text.split()
    return [" ".join(words[i:i+size]) for i in range(0, len(words), size - overlap)]

def create_faiss_index(chunks):
    vectors = embedder.encode(chunks)
    idx = faiss.IndexFlatL2(vectors.shape[1])
    idx.add(vectors)
    return idx

def retrieve_chunks(query, chunks, idx, k=3):
    q_vec = embedder.encode([query])
    _, indices = idx.search(q_vec, k)
    return [chunks[i] for i in indices[0]]

def build_prompt(query, context_chunks):
    context = "\n\n".join(context_chunks)
    return f"""You are a helpful assistant. Use the context below to answer the user's question.

Context:
{context}

Question:
{query}

Answer:"""

def llm_answer(prompt):
    return llm(prompt, max_new_tokens=256)

def process_pdf(file):
    global doc_chunks, index
    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
        tmp.write(file.read())
        text = extract_text_from_pdf(tmp.name)
    doc_chunks = chunk_text(text)
    index = create_faiss_index(doc_chunks)
    return "✅ PDF processed. Ask me anything!"

def chat_with_pdf(message):
    if not doc_chunks:
        return "❌ Upload a PDF first."
    chunks = retrieve_chunks(message, doc_chunks, index)
    prompt = build_prompt(message, chunks)
    return llm_answer(prompt)

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## 🧠 Chat with your PDF (Open Source, No Login!)")
    file_input = gr.File(label="Upload PDF")
    upload_btn = gr.Button("Process PDF")
    chatbot = gr.ChatInterface(fn=chat_with_pdf, textbox=gr.Textbox(placeholder="Ask something from your PDF..."))

    upload_btn.click(fn=process_pdf, inputs=[file_input], outputs=[chatbot.textbox])

demo.launch()