import os import time import hashlib import json from dotenv import load_dotenv import streamlit as st from langchain_community.vectorstores import FAISS from langchain_community.embeddings import HuggingFaceEmbeddings from langchain.prompts import PromptTemplate from langchain_together import Together from langchain.memory import ConversationBufferWindowMemory from langchain.chains import ConversationalRetrievalChain from PyPDF2 import PdfReader, PdfWriter from io import BytesIO from reportlab.pdfgen import canvas from reportlab.graphics.barcode import code128 from reportlab.lib.pagesizes import letter from reportlab.lib.units import mm load_dotenv() st.set_page_config(page_title="LawGPT", layout="wide") st.markdown(""" """, unsafe_allow_html=True) st.markdown("""
""", unsafe_allow_html=True) if "role" not in st.session_state: st.session_state.role = None if "authenticated" not in st.session_state: st.session_state.authenticated = False if st.session_state.role is None: st.markdown("

Who are you?

", unsafe_allow_html=True) col1, col2, col3 = st.columns([1, 3, 1]) with col2: col_a, col_b, col_c = st.columns(3) with col_a: if st.button("🧑 I am a Civilian"): st.session_state.role = "civilian" st.session_state.authenticated = True st.rerun() with col_b: if st.button("⚖ī¸ I am a Court Stakeholder"): st.session_state.role = "stakeholder" st.rerun() with col_c: if st.button("👨‍⚖ī¸ I am a Judge"): st.session_state.role = "judge" st.rerun() if (st.session_state.role == "stakeholder" or st.session_state.role == "judge") and not st.session_state.authenticated: st.markdown(f"### 🔐 {'Judge' if st.session_state.role == 'judge' else 'Stakeholder'} Login") username = st.text_input("Username") password = st.text_input("Password", type="password") if st.button("Login"): if username == "admin" and password == "1234": st.success("Login successful!") st.session_state.authenticated = True st.rerun() else: st.error("Invalid credentials.") if st.session_state.role and (st.session_state.role == "civilian" or st.session_state.authenticated): if st.button("🔙 Back to Home"): st.session_state.role = None st.session_state.authenticated = False st.rerun() tabs = ["📘 LawGPT"] if st.session_state.role == "judge": tabs.extend(["👨‍⚖ī¸ Judge Console", "📜 Previous Judgments"]) elif st.session_state.role == "stakeholder": tabs.extend(["📝 Document Signer", "🔍 Verify Document"]) selected_tab = st.tabs(tabs) # Load embeddings and DB for all roles embeddings = HuggingFaceEmbeddings( model_name="nomic-ai/nomic-embed-text-v1", model_kwargs={"trust_remote_code": True, "revision": "289f532e14dbbbd5a04753fa58739e9ba766f3c7"} ) db = FAISS.load_local("ipc_vector_db", embeddings, allow_dangerous_deserialization=True) db_retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": 4}) # Common LLM setup llm = Together( model="mistralai/Mistral-7B-Instruct-v0.2", temperature=0.5, max_tokens=1024, together_api_key=os.getenv("TOGETHER_API_KEY") ) # LawGPT Tab for all roles if "📘 LawGPT" in tabs: with selected_tab[0]: st.markdown("## đŸ’Ŧ Your Legal AI Lawyer") st.markdown("### Ask any legal question related to the Indian Penal Code (IPC)") st.markdown("Questions might be of types like: Suppose a 16 year old is drinking and driving, and hit a pedestrian on the road. What are the possible case laws imposed and give any one previous court decisions on the same.") def reset_conversation(): st.session_state.messages = [] st.session_state.memory.clear() if "messages" not in st.session_state: st.session_state.messages = [] if "memory" not in st.session_state: st.session_state.memory = ConversationBufferWindowMemory( k=2, memory_key="chat_history", return_messages=True ) prompt_template = """[INST]You are a legal chatbot that answers questions about the Indian Penal Code (IPC). Provide clear, concise, and accurate responses based on context and user's question. Avoid extra details or assumptions. Focus only on legal information. CONTEXT: {context} CHAT HISTORY: {chat_history} QUESTION: {question} ANSWER: [INST]""" prompt = PromptTemplate( template=prompt_template, input_variables=["context", "question", "chat_history"] ) qa = ConversationalRetrievalChain.from_llm( llm=llm, memory=st.session_state.memory, retriever=db_retriever, combine_docs_chain_kwargs={ 'prompt': prompt, 'document_variable_name': 'context' } ) chat_placeholder = st.empty() with chat_placeholder.container(): for msg in st.session_state.messages: with st.chat_message(msg["role"]): st.write(msg["content"]) input_prompt = st.chat_input("Ask a legal question...") if input_prompt: with st.chat_message("user"): st.write(input_prompt) st.session_state.messages.append({"role": "user", "content": input_prompt}) with st.chat_message("assistant"): with st.status("Thinking 💡", expanded=True): result = qa.invoke(input=input_prompt) message_placeholder = st.empty() full_response = "⚠ī¸ **_Note: Information provided may be inaccurate._**\n\n" for chunk in result["answer"]: full_response += chunk time.sleep(0.02) message_placeholder.markdown(full_response + " ▌") message_placeholder.markdown(full_response) st.session_state.messages.append({"role": "assistant", "content": result["answer"]}) st.button("🔄 Reset Chat", on_click=reset_conversation) # Judge Console Tab if st.session_state.role == "judge": # Initialize judgment storage if "judgments" not in st.session_state: st.session_state.judgments = [] # Load existing judgments if file exists try: with open("judgments.json", "r") as f: st.session_state.judgments = json.load(f) except (FileNotFoundError, json.JSONDecodeError): pass with selected_tab[1]: st.markdown("## 👨‍⚖ī¸ Judge's Decision Console") st.markdown("### Enter case details for analysis and judgment") # Input fields for case details st.subheader("Case Information") case_number = st.text_input("Case Number/ID") case_title = st.text_input("Case Title") plaintiff = st.text_input("Plaintiff/Prosecution") defendant = st.text_input("Defendant/Accused") # Case facts and context st.subheader("Case Details") case_facts = st.text_area("Enter detailed facts of the case:", height=200) # Get relevant laws/sections that apply relevant_laws = st.text_area("Relevant IPC Sections (if known):", placeholder="e.g. Section 302, Section 376, etc.") col1, col2 = st.columns(2) with col1: case_type = st.selectbox("Case Type", [ "Criminal", "Civil", "Family", "Property", "Cyber Crime", "Corporate", "Intellectual Property", "Other" ]) with col2: case_priority = st.select_slider("Case Priority", options=["Low", "Medium", "High", "Urgent"]) if st.button("Generate Judgment"): if not case_facts: st.error("Please enter the case facts to generate a judgment.") else: with st.status("Analyzing case and formulating judgment...", expanded=True): # Create a prompt for legal judgment judge_prompt_template = """[INST]You are an experienced Indian judge making a legal judgment based on the Indian Penal Code (IPC). Review the case details and provide a comprehensive legal judgment. CASE NUMBER: {case_number} CASE TITLE: {case_title} PLAINTIFF/PROSECUTION: {plaintiff} DEFENDANT/ACCUSED: {defendant} CASE TYPE: {case_type} CASE FACTS: {case_facts} RELEVANT IPC SECTIONS: {relevant_laws} Your judgment should follow this structure: 1. Summary of the case 2. Facts of the case 3. Legal issues involved 4. Analysis of applicable laws and precedents 5. Reasoning and findings 6. Final judgment and orders 7. Any remedies or penalties imposed Be impartial, consider only facts and relevant laws, and make a fair judgment. [INST]""" judge_prompt = PromptTemplate( template=judge_prompt_template, input_variables=["case_number", "case_title", "plaintiff", "defendant", "case_type", "case_facts", "relevant_laws"] ) # Format prompt with case details formatted_prompt = judge_prompt.format( case_number=case_number if case_number else "Unassigned", case_title=case_title if case_title else "Unnamed Case", plaintiff=plaintiff if plaintiff else "Unspecified", defendant=defendant if defendant else "Unspecified", case_type=case_type, case_facts=case_facts, relevant_laws=relevant_laws if relevant_laws else "To be determined" ) # Generate judgment using LLM judgment_result = llm.invoke(formatted_prompt) # Save judgment to session state timestamp = time.strftime("%Y-%m-%d %H:%M:%S") judgment_data = { "id": hashlib.md5(f"{case_title}{timestamp}".encode()).hexdigest()[:8], "case_number": case_number if case_number else "Unassigned", "case_title": case_title if case_title else "Unnamed Case", "plaintiff": plaintiff, "defendant": defendant, "case_type": case_type, "priority": case_priority, "facts": case_facts, "relevant_laws": relevant_laws, "judgment": judgment_result, "timestamp": timestamp } st.session_state.judgments.append(judgment_data) # Save judgments to file with open("judgments.json", "w") as f: json.dump(st.session_state.judgments, f) # Display the judgment st.markdown("### Judgment Generated") with st.container(): st.markdown(f"
", unsafe_allow_html=True) st.markdown(f"
JUDGMENT #{judgment_data['id']}
", unsafe_allow_html=True) st.markdown(f"**Case**: {judgment_data['case_title']}") st.markdown(f"**Date**: {judgment_data['timestamp']}") st.markdown("---") st.markdown(judgment_data['judgment']) st.markdown("
", unsafe_allow_html=True) # Download judgment as PDF if st.button("đŸ“Ĩ Download Judgment as PDF"): # Generate PDF with ReportLab pdf_buffer = BytesIO() c = canvas.Canvas(pdf_buffer, pagesize=letter) width, height = letter # Header c.setFont("Helvetica-Bold", 16) c.drawString(72, height - 72, f"JUDGMENT #{judgment_data['id']}") # Case details c.setFont("Helvetica-Bold", 12) c.drawString(72, height - 100, f"Case: {judgment_data['case_title']}") c.setFont("Helvetica", 10) c.drawString(72, height - 115, f"Case Number: {judgment_data['case_number']}") c.drawString(72, height - 130, f"Date: {judgment_data['timestamp']}") c.drawString(72, height - 145, f"Plaintiff/Prosecution: {judgment_data['plaintiff']}") c.drawString(72, height - 160, f"Defendant/Accused: {judgment_data['defendant']}") c.drawString(72, height - 175, f"Case Type: {judgment_data['case_type']}") # Line separator c.line(72, height - 190, width - 72, height - 190) # Format judgment text judgment_text = judgment_data['judgment'] text_object = c.beginText(72, height - 210) text_object.setFont("Times-Roman", 10) # Wrap text to fit page lines = [] for paragraph in judgment_text.split('\n\n'): # Replace single newlines with spaces for proper wrapping paragraph = paragraph.replace('\n', ' ') # Simple word wrap words = paragraph.split() line = '' for word in words: if len(line + ' ' + word) <= 90: # character limit per line line += ' ' + word if line else word else: lines.append(line) line = word if line: lines.append(line) # Add blank line between paragraphs lines.append('') # Add lines to text object with pagination line_height = 12 lines_per_page = 50 current_line = 0 for line in lines: if current_line >= lines_per_page: c.drawText(text_object) c.showPage() text_object = c.beginText(72, height - 72) text_object.setFont("Times-Roman", 10) current_line = 0 text_object.textLine(line) current_line += 1 c.drawText(text_object) # Add footer with page numbers c.saveState() c.setFont("Helvetica", 8) c.drawString(width/2 - 40, 30, f"Generated by LawGPT Judge") c.restoreState() # Final page with signature c.showPage() c.setFont("Helvetica-Bold", 12) c.drawString(72, height - 100, "OFFICIAL JUDGMENT") c.setFont("Helvetica", 10) c.drawString(72, height - 130, f"Case #{judgment_data['id']} - {judgment_data['case_title']}") c.drawString(72, height - 150, f"Date: {judgment_data['timestamp']}") # Add barcode for authenticity barcode = code128.Code128(f"JUDGMENT-{judgment_data['id']}", barHeight=10 * mm, barWidth=0.4) barcode.drawOn(c, 72, 100) # Add signature line c.line(width - 200, 70, width - 72, 70) c.drawString(width - 180, 60, "Judge's Signature") c.save() pdf_buffer.seek(0) # Offer download st.download_button( label="đŸ“Ĩ Download Generated PDF", data=pdf_buffer, file_name=f"judgment_{judgment_data['id']}_{judgment_data['case_title'].replace(' ', '_')}.pdf", mime="application/pdf" ) st.success("Judgment has been saved to the system.") # Previous Judgments Tab with selected_tab[2]: st.markdown("## 📜 Previous Judgments") st.markdown("### Review and search past judgments") # Search and filter search_term = st.text_input("Search judgments:", placeholder="Enter case title, number, plaintiff, etc.") col1, col2 = st.columns(2) with col1: filter_type = st.multiselect("Filter by case type:", options=["All"] + ["Criminal", "Civil", "Family", "Property", "Cyber Crime", "Corporate", "Intellectual Property", "Other"], default=["All"]) with col2: sort_by = st.selectbox("Sort by:", options=["Most recent", "Oldest first", "Case title (A-Z)"]) # Display judgments based on filters if len(st.session_state.judgments) == 0: st.info("No judgments recorded yet. Use the Judge Console to create judgments.") else: # Filter judgments filtered_judgments = st.session_state.judgments # Apply search term filter if search_term: filtered_judgments = [j for j in filtered_judgments if search_term.lower() in j['case_title'].lower() or search_term.lower() in j['case_number'].lower() or search_term.lower() in j['plaintiff'].lower() or search_term.lower() in j['defendant'].lower() or search_term.lower() in j.get('relevant_laws', '').lower()] # Apply case type filter if "All" not in filter_type: filtered_judgments = [j for j in filtered_judgments if j['case_type'] in filter_type] # Apply sorting if sort_by == "Most recent": filtered_judgments = sorted(filtered_judgments, key=lambda x: x['timestamp'], reverse=True) elif sort_by == "Oldest first": filtered_judgments = sorted(filtered_judgments, key=lambda x: x['timestamp']) elif sort_by == "Case title (A-Z)": filtered_judgments = sorted(filtered_judgments, key=lambda x: x['case_title']) # Display judgments for judgment in filtered_judgments: with st.expander(f"**{judgment['case_title']}** - {judgment['timestamp']}"): st.markdown(f"
JUDGMENT #{judgment['id']}
", unsafe_allow_html=True) st.markdown(f"**Case Number**: {judgment['case_number']}") st.markdown(f"**Plaintiff**: {judgment['plaintiff']}") st.markdown(f"**Defendant**: {judgment['defendant']}") st.markdown(f"**Case Type**: {judgment['case_type']} (Priority: {judgment['priority']})") st.markdown("#### Case Facts") st.markdown(judgment['facts']) if judgment.get('relevant_laws'): st.markdown("#### Relevant Laws Applied") st.markdown(judgment['relevant_laws']) st.markdown("#### Full Judgment") st.markdown("---") st.markdown(judgment['judgment']) # Button to download individual judgment as PDF if st.button(f"đŸ“Ĩ Download PDF", key=f"download_{judgment['id']}"): # Generate PDF with ReportLab pdf_buffer = BytesIO() c = canvas.Canvas(pdf_buffer, pagesize=letter) width, height = letter # Header c.setFont("Helvetica-Bold", 16) c.drawString(72, height - 72, f"JUDGMENT #{judgment['id']}") # Case details c.setFont("Helvetica-Bold", 12) c.drawString(72, height - 100, f"Case: {judgment['case_title']}") c.setFont("Helvetica", 10) c.drawString(72, height - 115, f"Case Number: {judgment['case_number']}") c.drawString(72, height - 130, f"Date: {judgment['timestamp']}") c.drawString(72, height - 145, f"Plaintiff/Prosecution: {judgment['plaintiff']}") c.drawString(72, height - 160, f"Defendant/Accused: {judgment['defendant']}") c.drawString(72, height - 175, f"Case Type: {judgment['case_type']}") # Line separator c.line(72, height - 190, width - 72, height - 190) # Format judgment text judgment_text = judgment['judgment'] text_object = c.beginText(72, height - 210) text_object.setFont("Times-Roman", 10) # Wrap text to fit page lines = [] for paragraph in judgment_text.split('\n\n'): # Replace single newlines with spaces for proper wrapping paragraph = paragraph.replace('\n', ' ') # Simple word wrap words = paragraph.split() line = '' for word in words: if len(line + ' ' + word) <= 90: # character limit per line line += ' ' + word if line else word else: lines.append(line) line = word if line: lines.append(line) # Add blank line between paragraphs lines.append('') # Add lines to text object with pagination line_height = 12 lines_per_page = 50 current_line = 0 for line in lines: if current_line >= lines_per_page: c.drawText(text_object) c.showPage() text_object = c.beginText(72, height - 72) text_object.setFont("Times-Roman", 10) current_line = 0 text_object.textLine(line) current_line += 1 c.drawText(text_object) # Add footer with page numbers c.saveState() c.setFont("Helvetica", 8) c.drawString(width/2 - 40, 30, f"Generated by LawGPT Judge") c.restoreState() # Final page with signature c.showPage() c.setFont("Helvetica-Bold", 12) c.drawString(72, height - 100, "OFFICIAL JUDGMENT") c.setFont("Helvetica", 10) c.drawString(72, height - 130, f"Case #{judgment['id']} - {judgment['case_title']}") c.drawString(72, height - 150, f"Date: {judgment['timestamp']}") # Add barcode for authenticity barcode = code128.Code128(f"JUDGMENT-{judgment['id']}", barHeight=10 * mm, barWidth=0.4) barcode.drawOn(c, 72, 100) # Add signature line c.line(width - 200, 70, width - 72, 70) c.drawString(width - 180, 60, "Judge's Signature") c.save() pdf_buffer.seek(0) # Offer download st.download_button( label="đŸ“Ĩ Download Generated PDF", data=pdf_buffer, file_name=f"judgment_{judgment['id']}_{judgment['case_title'].replace(' ', '_')}.pdf", mime="application/pdf", key=f"pdf_{judgment['id']}" ) # Stakeholder tabs if st.session_state.role == "stakeholder": if "📝 Document Signer" in tabs: with selected_tab[1]: st.markdown("## 📝 Upload and Sign Document") uploaded_file = st.file_uploader("Choose a file to sign", type=["pdf"]) signer_name = st.text_input("Enter your name (Signer):") if uploaded_file and signer_name: file_content = uploaded_file.read() input_pdf = BytesIO(file_content) output_pdf = BytesIO() reader = PdfReader(input_pdf) writer = PdfWriter() for page in reader.pages: page_width = float(page.mediabox.width) page_height = float(page.mediabox.height) packet = BytesIO() can = canvas.Canvas(packet, pagesize=(page_width, page_height)) barcode = code128.Code128(signer_name, barHeight=10 * mm, barWidth=0.4) barcode.drawOn(can, 50, 50) can.setFont("Helvetica", 10) can.drawString(50, 40, f"Signed by: {signer_name}") can.save() packet.seek(0) overlay = PdfReader(packet).pages[0] page.merge_page(overlay) writer.add_page(page) writer.write(output_pdf) output_pdf.seek(0) st.download_button("📅 Download Signed Document", output_pdf, file_name=f"signed_{uploaded_file.name}", mime="application/pdf") if "🔍 Verify Document" in tabs: with selected_tab[2]: st.markdown("## 🔍 Verify Document Authentication") st.markdown("Upload any document to verify its integrity and authenticity.") verify_file = st.file_uploader("Upload PDF for verification", type=["pdf"], key="verify") if verify_file: content = verify_file.read() try: # Basic PDF validation pdf = PdfReader(BytesIO(content)) # Extract text to look for signature markers all_text = "" for page in pdf.pages: all_text += page.extract_text() or "" # Check for digital signature information has_signature_text = any(sig_text in all_text.lower() for sig_text in ["signed by:", "digital signature", "electronic signature"]) # Create document fingerprint/hash doc_hash = hashlib.sha256(content).hexdigest() # Calculate metadata integrity metadata_valid = True if pdf.metadata: try: # Check for suspicious metadata modifications creation_date = pdf.metadata.get('/CreationDate', '') mod_date = pdf.metadata.get('/ModDate', '') if mod_date and creation_date: metadata_valid = mod_date >= creation_date except: metadata_valid = False # Check for content consistency content_consistent = True col1, col2 = st.columns(2) with col1: st.subheader("Document Analysis") st.info(f"📄 Pages: {len(pdf.pages)}") st.info(f"🔒 Contains signature markers: {'Yes' if has_signature_text else 'No'}") # Display hash for document tracking st.code(f"Document Hash: {doc_hash[:16]}...{doc_hash[-16:]}") # Document size and characteristics file_size = len(content) / 1024 # KB st.info(f"đŸ“Ļ File size: {file_size:.2f} KB") with col2: st.subheader("Verification Results") # Case 1: Document has signature markers if has_signature_text: if metadata_valid and content_consistent: st.success("✅ Document Status: VERIFIED AUTHENTIC") st.markdown("- ✓ Valid PDF structure") st.markdown("- ✓ Signature information detected") st.markdown("- ✓ No tampering indicators found") st.markdown("- ✓ Metadata consistency verified") else: st.warning("⚠ī¸ Document Status: POTENTIALLY MODIFIED") st.markdown("- ✓ Valid PDF structure") st.markdown("- ✓ Signature information found") st.markdown("- ❌ Some integrity checks failed") if not metadata_valid: st.markdown("- ❌ Metadata inconsistencies detected") # Display signature extraction if present signature_line = next((line for line in all_text.split('\n') if "signed by:" in line.lower()), "") if signature_line: st.info(f"📝 {signature_line.strip()}") # Case 2: Document without signatures else: if metadata_valid and content_consistent: st.success("✅ Document Status: VALID DOCUMENT") st.markdown("- ✓ Valid PDF structure") st.markdown("- ✓ Content integrity verified") st.markdown("- ✓ No tampering indicators found") st.markdown("- ℹī¸ No signature information found (this is not an error)") else: st.warning("⚠ī¸ Document Status: POTENTIALLY MODIFIED") st.markdown("- ✓ Valid PDF structure") st.markdown("- ❌ Some integrity checks failed") if not metadata_valid: st.markdown("- ❌ Metadata inconsistencies detected") # Advanced options with st.expander("đŸ”Ŧ Advanced Verification Details"): st.markdown("### Document Metadata") if pdf.metadata: for key, value in pdf.metadata.items(): if key and value and key not in ('/CreationDate', '/ModDate'): st.text(f"{key}: {value}") else: st.text("No metadata available") st.markdown("### Integrity Timeline") st.text(f"Creation Date: {pdf.metadata.get('/CreationDate', 'Not available')}") st.text(f"Last Modified: {pdf.metadata.get('/ModDate', 'Not available')}") # Additional verification for content integrity st.markdown("### Content Analysis") fonts_used = set() image_count = 0 for page in pdf.pages: if "/Font" in page["/Resources"]: for font in page["/Resources"]["/Font"]: fonts_used.add(str(font)) if "/XObject" in page["/Resources"]: for obj in page["/Resources"]["/XObject"]: if "/Subtype" in page["/Resources"]["/XObject"][obj] and \ page["/Resources"]["/XObject"][obj]["/Subtype"] == "/Image": image_count += 1 st.text(f"Fonts detected: {len(fonts_used)}") st.text(f"Images detected: {image_count}") except Exception as e: st.error(f"❌ Document Status: INVALID OR CORRUPTED") st.markdown(f"Error: Could not process the document properly. The file may be corrupted or not a valid PDF.") st.markdown(f"Technical details: {str(e)}")