Update app.py

app.py

@@ -1,97 +1,104 @@
-!pip install -q gradio sentence-transformers faiss-cpu pdfplumber groq
-
-import gradio as gr
-import pdfplumber
-from sentence_transformers import SentenceTransformer
+import os
+import streamlit as st
+from PyPDF2 import PdfReader
+from docx import Document
 import faiss
 import numpy as np
+import torch
 from groq import Groq
+from sentence_transformers import SentenceTransformer
 
-# Global vars
-model = SentenceTransformer("all-MiniLM-L6-v2")
-documents, embeddings, index, text_chunks, client = [], None, None, [], None
-
-def ask_llama3(system_prompt, user_prompt):
-    global client
-    try:
-        chat_completion = client.chat.completions.create(
-            model="llama-3.1-8b-instant",
-            messages=[
-                {"role": "system", "content": system_prompt},
-                {"role": "user", "content": user_prompt}
-            ]
-        )
-        return chat_completion.choices[0].message.content
-    except Exception as e:
-        return f"❌ LLaMA3 error: {e}"
-
-def pdf_to_chunks(pdf_file, user_key):
-    global text_chunks, embeddings, index, client
-    try:
-        client = Groq(api_key=user_key)
-    except Exception as e:
-        return f"❌ API key error: {e}"
-
-    text_chunks = []
-    try:
-        with pdfplumber.open(pdf_file.name) as pdf:
-            for page in pdf.pages:
-                text = page.extract_text()
-                if text:
-                    sentences = text.split(". ")
-                    text_chunks.extend(sentences)
-
-        if not text_chunks:
-            return "❗ No text found in PDF."
-
-        embeddings = model.encode(text_chunks, convert_to_tensor=False)
-        embeddings = np.array(embeddings).astype("float32")
-        dimension = embeddings.shape[1]
-        index = faiss.IndexFlatL2(dimension)
-        index.add(embeddings)
-
-        return "✅ PDF processed and indexed successfully."
-    except Exception as e:
-        return f"❌ PDF processing error: {e}"
-
-def query_document(question):
-    global index, text_chunks, model
-    if index is None or not text_chunks:
-        return "❗ Please upload and process a PDF first."
-
-    try:
-        q_embedding = model.encode([question])[0].astype("float32")
-        D, I = index.search(np.array([q_embedding]), 5)
-        retrieved_chunks = [text_chunks[i] for i in I[0]]
-        context = "\n".join(retrieved_chunks)
-
-        system_prompt = "You are a helpful study supervisor. Use the provided context to answer clearly."
-        user_prompt = f"Context:\n{context}\n\nQuestion:\n{question}"
-
-        return ask_llama3(system_prompt, user_prompt)
-    except Exception as e:
-        return f"❌ Query error: {e}"
-
-# UI
-with gr.Blocks() as demo:
-    gr.Markdown("""
-    <div style="text-align:center; background:#f97316; color:white; padding: 12px; border-radius: 10px;">
-        <h2>📘 PDF Study Assistant</h2>
-        <p>Ask questions from your uploaded PDF using Groq + LLaMA3</p>
-    </div>
-    """)
-
-    with gr.Column():
-        api_input = gr.Textbox(label="🔑 Groq API Key", type="password")
-        pdf_input = gr.File(label="📄 Upload PDF", file_types=[".pdf"])
-        upload_btn = gr.Button("📥 Extract & Index PDF", variant="primary")
-        status_output = gr.Textbox(label="🛠️ Status", interactive=False)
-
-        question = gr.Textbox(label="❓ Ask a Question", lines=2)
-        get_answer_btn = gr.Button("💬 Get Answer")
-        answer_output = gr.Textbox(label="📢 Answer", lines=10, interactive=False)
-
-    upload_btn.click(fn=pdf_to_chunks, inputs=[pdf_input, api_input], outputs=[status_output])
-    get_answer_btn.click(fn=query_document, inputs=[question], outputs=[answer_output])
-
-demo.launch()
+# ✅ Ensure model uses CPU to avoid meta tensor issue on Hugging Face
+torch.set_default_device("cpu")
+
+# ✅ Load Groq API key from environment variable
+client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
+
+# ✅ Load SentenceTransformer model safely
+embed_model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2", trust_remote_code=True)
+
+# ✅ FAISS index and chunk store
+INDEX = faiss.IndexFlatL2(384)
+stored_chunks = []
+
+# ✅ UI Styling for Streamlit
+st.markdown("""
+    <style>
+    .main-title {
+        font-size: 40px;
+        color: #2E86C1;
+        font-weight: bold;
+        text-align: center;
+        margin-bottom: 30px;
+    }
+    .card {
+        background-color: #ffffff;
+        padding: 20px;
+        border-radius: 15px;
+        box-shadow: 0 4px 10px rgba(0, 0, 0, 0.1);
+        margin-top: 20px;
+    }
+    body {
+        background-color: #f8fbfd;
+    }
+    </style>
+""", unsafe_allow_html=True)
+
+st.markdown('<div class="main-title">📄 Smart RAG Document QA Assistant</div>', unsafe_allow_html=True)
+
+# ✅ Extract text from files
+def extract_text(file):
+    if file.type == "application/pdf":
+        reader = PdfReader(file)
+        return " ".join([page.extract_text() or "" for page in reader.pages])
+    elif file.type == "application/vnd.openxmlformats-officedocument.wordprocessingml.document":
+        doc = Document(file)
+        return "\n".join([p.text for p in doc.paragraphs])
+    elif file.type.startswith("text"):
+        return file.read().decode("utf-8")
+    return ""
+
+# ✅ Split long text into small chunks
+def chunk_text(text, chunk_size=200):
+    words = text.split()
+    return [" ".join(words[i:i+chunk_size]) for i in range(0, len(words), chunk_size)]
+
+# ✅ Store chunks and their embeddings in FAISS
+def store_embeddings(chunks):
+    vectors = embed_model.encode(chunks)
+    INDEX.add(np.array(vectors, dtype=np.float32))
+    stored_chunks.extend(chunks)
+
+# ✅ Retrieve most relevant chunks for a given query
+def retrieve_similar_chunks(query, top_k=3):
+    query_vector = embed_model.encode([query])
+    distances, indices = INDEX.search(np.array(query_vector, dtype=np.float32), top_k)
+    return [stored_chunks[i] for i in indices[0]]
+
+# ✅ Ask Groq LLM to answer based on document context
+def get_llm_answer(query, context):
+    prompt = f"Answer the question based on the following context:\n\n{context}\n\nQuestion: {query}"
+    chat_completion = client.chat.completions.create(
+        messages=[{"role": "user", "content": prompt}],
+        model="llama3-70b-8192"
+    )
+    return chat_completion.choices[0].message.content
+
+# ✅ File upload and query UI
+uploaded_file = st.file_uploader("📁 Upload your document", type=["pdf", "docx", "txt"])
+query = st.text_input("💬 Ask a question about your document")
+
+if uploaded_file:
+    with st.spinner("Processing file..."):
+        text = extract_text(uploaded_file)
+        chunks = chunk_text(text)
+        store_embeddings(chunks)
+    st.success("✅ Document uploaded and indexed!")
+
+if st.button("🧠 Get Answer") and query:
+    with st.spinner("Thinking..."):
+        context = "\n\n".join(retrieve_similar_chunks(query))
+        answer = get_llm_answer(query, context)
+        st.markdown(f'<div class="card"><b>Answer:</b><br>{answer}</div>', unsafe_allow_html=True)
+
+st.markdown("<br><center style='color: grey;'>Built by Muqadas with ❤️ using Streamlit + Groq + FAISS</center>", unsafe_allow_html=True)