Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import requests
|
| 3 |
+
import warnings
|
| 4 |
+
import gradio as gr
|
| 5 |
+
from groq import Groq
|
| 6 |
+
from langchain_community.document_loaders import PyPDFLoader
|
| 7 |
+
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
| 8 |
+
from langchain_huggingface import HuggingFaceEmbeddings
|
| 9 |
+
from langchain_community.vectorstores import FAISS
|
| 10 |
+
|
| 11 |
+
# Suppress technical warnings for a clean logs
|
| 12 |
+
warnings.filterwarnings("ignore")
|
| 13 |
+
|
| 14 |
+
# --- 1. CONFIGURATION & SECRETS ---
|
| 15 |
+
# On Hugging Face, set 'MY_GROQ_SECRET' in Settings > Variables and Secrets
|
| 16 |
+
GROQ_API_KEY = os.environ.get("MY_GROQ_SECRET")
|
| 17 |
+
client = Groq(api_key=GROQ_API_KEY)
|
| 18 |
+
|
| 19 |
+
# YOUR HIDDEN LINKS (Never shown in UI)
|
| 20 |
+
GDRIVE_LINKS = [
|
| 21 |
+
"https://drive.google.com/file/d/10D3uJqBYG9gMWsNHcpTW4I6BKmA2otfH/view?usp=sharing"
|
| 22 |
+
]
|
| 23 |
+
|
| 24 |
+
# --- 2. KNOWLEDGE BASE INITIALIZATION ---
|
| 25 |
+
def download_gdrive_pdf(url, output_path):
|
| 26 |
+
try:
|
| 27 |
+
file_id = url.split('/')[-2]
|
| 28 |
+
download_url = f'https://drive.google.com/uc?export=download&id={file_id}'
|
| 29 |
+
response = requests.get(download_url)
|
| 30 |
+
if response.status_code == 200:
|
| 31 |
+
with open(output_path, 'wb') as f:
|
| 32 |
+
f.write(response.content)
|
| 33 |
+
return True
|
| 34 |
+
except:
|
| 35 |
+
return False
|
| 36 |
+
return False
|
| 37 |
+
|
| 38 |
+
# Initialize the vector database on startup
|
| 39 |
+
all_chunks = []
|
| 40 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=800, chunk_overlap=150)
|
| 41 |
+
|
| 42 |
+
for i, link in enumerate(GDRIVE_LINKS):
|
| 43 |
+
filename = f"doc_{i}.pdf"
|
| 44 |
+
if download_gdrive_pdf(link, filename):
|
| 45 |
+
try:
|
| 46 |
+
loader = PyPDFLoader(filename)
|
| 47 |
+
all_chunks.extend(text_splitter.split_documents(loader.load()))
|
| 48 |
+
finally:
|
| 49 |
+
if os.path.exists(filename):
|
| 50 |
+
os.remove(filename)
|
| 51 |
+
|
| 52 |
+
embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
|
| 53 |
+
vector_db = FAISS.from_documents(all_chunks, embeddings)
|
| 54 |
+
|
| 55 |
+
# --- 3. STRICT RAG LOGIC ---
|
| 56 |
+
def respond(message, history):
|
| 57 |
+
# Retrieve relevant snippets
|
| 58 |
+
docs = vector_db.similarity_search(message, k=5)
|
| 59 |
+
context = "\n\n".join([doc.page_content for doc in docs])
|
| 60 |
+
|
| 61 |
+
# Strict instructions: No outside knowledge allowed
|
| 62 |
+
system_prompt = f"""
|
| 63 |
+
You are a professional Knowledge Assistant.
|
| 64 |
+
1. Answer ONLY using the context provided.
|
| 65 |
+
2. If the answer is NOT in the context, say: "Answer not found in provided documents."
|
| 66 |
+
3. Do not mention the context or the technical nature of the search.
|
| 67 |
+
|
| 68 |
+
CONTEXT:
|
| 69 |
+
{context}
|
| 70 |
+
"""
|
| 71 |
+
|
| 72 |
+
chat_completion = client.chat.completions.create(
|
| 73 |
+
messages=[
|
| 74 |
+
{"role": "system", "content": system_prompt},
|
| 75 |
+
{"role": "user", "content": message}
|
| 76 |
+
],
|
| 77 |
+
model="llama-3.3-70b-versatile",
|
| 78 |
+
temperature=0.1,
|
| 79 |
+
)
|
| 80 |
+
return chat_completion.choices[0].message.content
|
| 81 |
+
|
| 82 |
+
# --- 4. ATTRACTIVE MODERN UI ---
|
| 83 |
+
custom_css = """
|
| 84 |
+
body { background-color: #0f172a; }
|
| 85 |
+
.gradio-container { max-width: 850px !important; margin: auto; padding-top: 50px; }
|
| 86 |
+
#title-text { text-align: center; color: #38bdf8; font-weight: 800; margin-bottom: 5px; }
|
| 87 |
+
#desc-text { text-align: center; color: #94a3b8; margin-bottom: 25px; }
|
| 88 |
+
.chat-container { border-radius: 20px !important; border: 1px solid #334155 !important; background: #1e293b !important; box-shadow: 0 20px 50px rgba(0,0,0,0.4); }
|
| 89 |
+
.primary-btn { background: linear-gradient(135deg, #38bdf8, #818cf8) !important; border: none !important; color: white !important; }
|
| 90 |
+
footer { display: none !important; }
|
| 91 |
+
"""
|
| 92 |
+
|
| 93 |
+
with gr.Blocks(theme=gr.themes.Default(primary_hue="sky"), css=custom_css) as demo:
|
| 94 |
+
gr.HTML("<h1 id='title-text'>🌀 DocuVortex</h1>")
|
| 95 |
+
# REPLACE "User's Research AI" with your own name here!
|
| 96 |
+
gr.HTML("<p id='desc-text'>User's Research AI: Strict Document Knowledge Base</p>")
|
| 97 |
+
|
| 98 |
+
with gr.Column(elem_id="chat-container"):
|
| 99 |
+
gr.ChatInterface(
|
| 100 |
+
fn=respond,
|
| 101 |
+
chatbot=gr.Chatbot(height=550, bubble_full_width=False, show_label=False),
|
| 102 |
+
textbox=gr.Textbox(placeholder="Ask a question about the document...", container=False, scale=7),
|
| 103 |
+
submit_btn=gr.Button("Ask AI", variant="primary", elem_classes="primary-btn"),
|
| 104 |
+
retry_btn=None,
|
| 105 |
+
undo_btn=None,
|
| 106 |
+
clear_btn=gr.Button("New Chat", variant="secondary")
|
| 107 |
+
)
|
| 108 |
+
|
| 109 |
+
if __name__ == "__main__":
|
| 110 |
+
demo.launch()
|