import os
import gradio as gr
from PyPDF2 import PdfReader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain.chains.retrieval_qa.base import RetrievalQA
from langchain.prompts import PromptTemplate
from langchain_core.language_models.llms import LLM
from langchain_core.callbacks import CallbackManagerForLLMRun
from typing import Optional, List, Dict, Any
from dotenv import load_dotenv
from groq import Groq
import urllib.parse
import feedparser
from numpy import dot
from numpy.linalg import norm
# Load environment variables
load_dotenv()
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
# -----------------------------------------------------------
# GROQ WRAPPER
# -----------------------------------------------------------
class GroqWrapper(LLM):
client: Any
model_name: str = "llama-3.3-70b-versatile"
temperature: float = 0.7
@property
def _llm_type(self) -> str:
return "groq"
def _call(
self,
prompt: str,
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> str:
response = self.client.chat.completions.create(
model=self.model_name,
messages=[{"role": "user", "content": prompt}],
temperature=self.temperature,
)
return response.choices[0].message.content
# Globals
vectorstore = None
qa_chain = None
groq_llm = None
# -----------------------------------------------------------
# PROCESS PDF
# -----------------------------------------------------------
def upload_pdf(file):
global vectorstore, qa_chain, groq_llm
try:
# Initialize Groq LLM
if groq_llm is None:
groq_llm = GroqWrapper(client=Groq(api_key=GROQ_API_KEY))
# Extract text from PDF
text = "".join(page.extract_text() or "" for page in PdfReader(file).pages)
if not text.strip():
return "Error: No readable text found in PDF"
# Chunk the text
splitter = RecursiveCharacterTextSplitter(
chunk_size=1000,
chunk_overlap=150,
separators=["\n\n", "\n", ".", "?", "!"]
)
chunks = splitter.split_text(text)
# Create Vectorstore
embeddings = HuggingFaceEmbeddings(
model_name="sentence-transformers/msmarco-MiniLM-L-12-v3"
)
vectorstore = FAISS.from_texts(chunks, embeddings)
# --- CUSTOM REFINE PROMPTS ---
initial_prompt = PromptTemplate(
input_variables=["context", "question"],
template="""
You are an expert researcher.
Use ONLY the given context to answer the question.
If the answer is not in the context, say "I don't know".
Context:
{context}
Question: {question}
Initial Answer:
"""
)
refine_prompt = PromptTemplate(
input_variables=["context", "question", "existing_answer"],
template="""
We have an existing answer:
{existing_answer}
Using the additional context below, refine the answer.
Additional Context:
{context}
Question: {question}
Refined Answer:
"""
)
# --- BUILD QA CHAIN ---
qa_chain = RetrievalQA.from_chain_type(
llm=groq_llm,
retriever=vectorstore.as_retriever(),
chain_type="refine",
return_source_documents=True,
chain_type_kwargs={
"question_prompt": initial_prompt,
"refine_prompt": refine_prompt,
"document_variable_name": "context" # <-- ADD THIS LINE
}
)
return "PDF processed successfully!"
except Exception as e:
return f"Error: {str(e)}"
# -----------------------------------------------------------
# QUESTION ANSWERING
# -----------------------------------------------------------
def ask_question(query):
global qa_chain
if qa_chain is None:
return "Please upload a PDF first.", ""
try:
result = qa_chain({"query": query})
answer = result["result"]
# Format sources
sources = result.get("source_documents", [])
if sources:
source_text = "\n\n---\n".join(
f"Source {i+1}:\n{doc.page_content[:500]}..."
for i, doc in enumerate(sources)
)
else:
source_text = "No sources found."
return answer, source_text
except Exception as e:
return f"Error: {str(e)}", ""
# -----------------------------------------------------------
# SUMMARIZE PDF
# -----------------------------------------------------------
def summarize_pdf(num_points=6):
global groq_llm, vectorstore
if vectorstore is None:
return "Please upload a PDF first."
try:
docs = vectorstore.similarity_search("summary", k=5)
context = "\n\n".join([d.page_content for d in docs])
prompt = f"""
Summarize the research paper in {num_points} bullet points.
Make it clear, meaningful, and highlight key contributions.
Content:
{context}
Summary:
"""
if groq_llm is None:
groq_llm = GroqWrapper(client=Groq(api_key=GROQ_API_KEY))
return groq_llm(prompt).strip()
except Exception as e:
return f"Error: {str(e)}"
# -----------------------------------------------------------
# FIND SIMILAR PAPERS (arXiv)
# -----------------------------------------------------------
def extract_title(text):
# Take the first non-empty line as the title
for line in text.split("\n"):
line = line.strip()
if line:
return line
return "Research Paper" # fallback if empty
def find_similar_papers():
global vectorstore
if vectorstore is None:
return "Please upload a PDF first."
try:
# Get full PDF text from all chunks
docs = vectorstore.similarity_search("", k=30)
full_pdf_text = " ".join(d.page_content for d in docs)
if not full_pdf_text.strip():
return "PDF content too small."
# ----------------------------
# 1️⃣ Extract only the title
# ----------------------------
title = extract_title(full_pdf_text)
query_text = title # Use only the title for arXiv search
# ----------------------------
# 2️⃣ Search arXiv
# ----------------------------
encoded_query = urllib.parse.quote(query_text)
url = f"http://export.arxiv.org/api/query?search_query=all:{encoded_query}&start=0&max_results=15"
feed = feedparser.parse(url)
entries = feed.entries
if not entries:
return "No similar papers found on arXiv."
# ----------------------------
# 3️⃣ Use embeddings for ranking
# ----------------------------
embedding_model = HuggingFaceEmbeddings(
model_name="sentence-transformers/all-mpnet-base-v2"
)
query_emb = embedding_model.embed_query(query_text)
ranked = []
for entry in entries:
candidate_text = entry.title # only title for similarity
emb = embedding_model.embed_query(candidate_text)
sim = dot(query_emb, emb) / (norm(query_emb) * norm(emb))
ranked.append({
"title": entry.title,
"summary": entry.summary.replace("\n", " ").strip(),
"link": entry.link,
"similarity": sim
})
# Sort by similarity
ranked.sort(key=lambda x: x["similarity"], reverse=True)
# ----------------------------
# 4️⃣ Format top 3 results
# ----------------------------
output = []
for p in ranked[:3]:
out = (
f"**{p['title']}**\n"
f"{p['summary']}\n"
f"🔗 {p['link']}\n"
f"Similarity Score: {p['similarity']:.2f}"
)
output.append(out)
return "\n\n".join(output)
except Exception as e:
return f"Error: {str(e)}"
css = '''
:root {
--primary: #6e48aa;
--secondary: #9d50bb;
--accent: #4776e6;
--dark: #1a1a2e;
--darker: #16213e;
--light: #f8f9fa;
--success: #4caf50;
--warning: #ff9800;
--danger: #f44336;
}
body, .gradio-container {
margin: 0;
padding: 0;
font-family: 'Segoe UI', 'Roboto', sans-serif;
background: linear-gradient(135deg, var(--dark), var(--darker));
color: var(--light);
min-height: 100vh;
}
.header {
text-align: center;
padding: 1.5rem 0;
margin-bottom: 2rem;
color: white; /* Make text white */
font-size: 3rem;
font-weight: 800;
letter-spacing: 1px;
font-style: italic; /* Make it italic */
text-shadow: 0 2px 10px rgba(0,0,0,0.2);
}
.nav-tabs {
display: flex;
justify-content: center;
margin-bottom: 2rem;
gap: 1rem;
}
.tab-button {
background: rgba(255,255,255,0.1);
border: none;
padding: 0.8rem 1.5rem;
border-radius: 50px;
color: white;
font-weight: 600;
cursor: pointer;
transition: all 0.3s ease;
box-shadow: 0 4px 6px rgba(0,0,0,0.1);
}
.tab-button:hover {
background: rgba(255,255,255,0.2);
transform: translateY(-2px);
}
.tab-button.active {
background: linear-gradient(45deg, var(--primary), var(--accent));
box-shadow: 0 4px 15px rgba(110, 72, 170, 0.4);
}
.tab-content {
display: none;
animation: fadeIn 0.5s ease-out;
}
.tab-content.active {
display: block;
}
.panel {
background: rgba(255,255,255,0.05);
border-radius: 16px;
padding: 2rem;
margin: 1rem auto;
max-width: 900px;
backdrop-filter: blur(10px);
border: 1px solid rgba(255,255,255,0.1);
box-shadow: 0 8px 32px rgba(0,0,0,0.2);
}
.panel-header {
font-size: 1.5rem;
font-weight: 700;
margin-bottom: 1.5rem;
color: white;
display: flex;
align-items: center;
gap: 0.8rem;
}
.panel-header svg {
width: 1.5rem;
height: 1.5rem;
}
button {
background: linear-gradient(45deg, var(--primary), var(--secondary));
color: white;
border: none;
padding: 0.8rem 1.5rem;
border-radius: 50px;
font-weight: 600;
cursor: pointer;
transition: all 0.3s ease;
box-shadow: 0 4px 15px rgba(110, 72, 170, 0.3);
margin: 0.5rem 0;
}
button:hover {
transform: translateY(-2px);
box-shadow: 0 6px 20px rgba(110, 72, 170, 0.4);
}
button:active {
transform: translateY(0);
}
button.secondary {
background: rgba(255,255,255,0.1);
}
button.secondary:hover {
background: rgba(255,255,255,0.2);
}
textarea, input[type="text"] {
background: rgba(255,255,255,0.1);
border: 1px solid rgba(255,255,255,0.2);
color: white;
border-radius: 8px;
padding: 0.8rem;
width: 100%;
margin-bottom: 1rem;
}
textarea:focus, input[type="text"]:focus {
outline: none;
border-color: var(--accent);
box-shadow: 0 0 0 2px rgba(71, 118, 230, 0.3);
}
.output-box {
background: rgba(0,0,0,0.3);
border-radius: 8px;
padding: 1rem;
margin-top: 1rem;
border-left: 4px solid var(--accent);
}
.output-label {
font-weight: 600;
margin-bottom: 0.5rem;
display: block;
color: #ddd;
}
@keyframes fadeIn {
from { opacity: 0; transform: translateY(10px); }
to { opacity: 1; transform: translateY(0); }
}
.slide-in {
animation: slideIn 0.5s ease-out forwards;
}
@keyframes slideIn {
from { transform: translateX(100%); opacity: 0; }
to { transform: translateX(0); opacity: 1; }
}
.file-upload {
border: 2px dashed rgba(255,255,255,0.3);
border-radius: 8px;
padding: 2rem;
text-align: center;
margin-bottom: 1rem;
transition: all 0.3s ease;
}
.file-upload:hover {
border-color: var(--accent);
background: rgba(71, 118, 230, 0.1);
}
.progress-bar {
height: 6px;
background: rgba(255,255,255,0.1);
border-radius: 3px;
margin-top: 1rem;
overflow: hidden;
}
.progress {
height: 100%;
background: linear-gradient(90deg, var(--primary), var(--accent));
width: 0%;
transition: width 0.3s ease;
}
'''
with gr.Blocks(css=css) as demo:
gr.Markdown("""
""")
with gr.Tabs() as tabs:
with gr.TabItem("📄 Upload PDF", id="upload"):
with gr.Column(elem_classes=["panel"]):
gr.Markdown("""""")
with gr.Column(elem_classes=["file-upload"]):
file_upload = gr.File(
file_types=['.pdf'],
label="Drag & Drop PDF or Click to Browse",
elem_classes=["upload-box"]
)
upload_btn = gr.Button("Process Document", variant="primary")
status = gr.Textbox(label="Processing Status", interactive=False)
gr.Markdown("")
with gr.TabItem("❓ Ask Questions", id="qa"):
with gr.Column(elem_classes=["panel"]):
gr.Markdown("""""")
question = gr.Textbox(
placeholder="Type your research question here...",
label="Your Question",
lines=3
)
ask_btn = gr.Button("Get Answer", variant="primary")
with gr.Column(elem_classes=["output-box"]):
gr.Markdown("Answer
")
answer = gr.Textbox(show_label=False, lines=6, interactive=False)
with gr.Column(elem_classes=["output-box"]):
gr.Markdown("Source References
")
citations = gr.Textbox(show_label=False, lines=4, interactive=False)
with gr.TabItem("✍️ Summarize", id="summary"):
with gr.Column(elem_classes=["panel"]):
gr.Markdown("""""")
summary_btn = gr.Button("Generate Summary", variant="primary")
with gr.Column(elem_classes=["output-box"]):
gr.Markdown("Key Insights
")
summary_output = gr.Textbox(show_label=False, lines=8, interactive=False)
with gr.TabItem("🔍 Similar Papers", id="papers"):
with gr.Column(elem_classes=["panel"]):
gr.Markdown("""""")
similar_btn = gr.Button("Find Similar Papers", variant="primary")
with gr.Column(elem_classes=["output-box"]):
gr.Markdown("Recommended Papers
")
similar_output = gr.Textbox(show_label=False, lines=8, interactive=False)
# Event handlers
upload_btn.click(upload_pdf, inputs=file_upload, outputs=status)
ask_btn.click(ask_question, inputs=question, outputs=[answer, citations])
summary_btn.click(summarize_pdf, outputs=summary_output)
similar_btn.click(find_similar_papers, outputs=similar_output)
if __name__ == "__main__":
demo.launch()