#!/usr/bin/env python # coding=utf-8 import csv import datetime import os import re from fpdf import FPDF from langdetect import detect import gradio as gr from tools.common import prepend_metadata_questions # === PDF Export Function with Language Option === def export_text_to_pdf(text, answers, output_path=None, language="fr"): if output_path is None: timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S") output_path = f"ai_act_register_{timestamp}.pdf" pdf = FPDF() pdf.add_page() pdf.set_auto_page_break(auto=True, margin=15) # Title pdf.set_font("Arial", 'B', 16) pdf.set_text_color(0, 51, 102) title = "Documentation Record for High-Risk AI Systems" if language == "en" else "Registre de Conformité AI Act" pdf.cell(0, 15, title, ln=True, align='C') pdf.ln(5) # Metadata below title pdf.set_font("Arial", 'I', 11) pdf.set_text_color(80, 80, 80) name = answers.get("user_name", "N/A") role = answers.get("user_role", "N/A") org = answers.get("organization_name", "N/A") timestamp = datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S') pdf.multi_cell(0, 10, f"Completed by {name} ({role}) at {org} on {timestamp}", align="C") pdf.ln(5) # Content pdf.set_font("Arial", '', 12) pdf.set_text_color(0, 0, 0) for line in text.strip().split('\n'): line = line.strip() if line.startswith("## "): section_title = line.replace("## ", "").strip() pdf.set_font("Arial", 'B', 13) pdf.set_text_color(30, 30, 120) pdf.ln(8) pdf.cell(0, 10, section_title, ln=True) pdf.set_font("Arial", '', 12) pdf.set_text_color(0, 0, 0) elif line.startswith("- **"): match = re.match(r"- \*\*(.+?)\*\*: (.+)", line) if match: label, value = match.groups() pdf.set_font("Arial", 'B', 12) pdf.cell(0, 10, f"{label}:", ln=True) pdf.set_font("Arial", '', 12) pdf.multi_cell(0, 10, value) pdf.ln(2) elif line == "---": pdf.line(10, pdf.get_y(), 200, pdf.get_y()) pdf.ln(5) else: pdf.multi_cell(0, 10, line) pdf.ln(2) pdf.output(output_path) return output_path # === Sequential Questions === QUESTIONS = prepend_metadata_questions([ ("responsible_person", "Who is responsible for this AI system?"), ("deployment_date", "When is the AI system scheduled to be deployed?"), ("ai_type", "What type of AI system is it?"), ("ai_description", "Please briefly describe what the system does."), ("risk_level", "What is the risk level of this system (e.g., high, medium)?"), ("risk_justification", "Why do you consider it this risk level?"), ("data_evaluation", "How have you evaluated the training data?"), ("technical_docs", "What technical documentation is available?"), ("human_oversight", "What kind of human oversight is planned?"), ("transparency_measures", "What transparency mechanisms are in place?"), ("audit_frequency", "How often will the system be audited?"), ("compliance_contact", "Who is the contact person for compliance (email or name)?") ]) # === Interactive Collection Flow === def step_by_step_agent(user_input, state): if state is None: state = {"step": 0, "answers": {}} step = state["step"] answers = state["answers"] if step > 0: key, _ = QUESTIONS[step - 1] answers[key] = user_input if step < len(QUESTIONS): next_q = QUESTIONS[step][1] state["step"] += 1 return next_q, state, None # Build filled template filled = f""" # AI Act Compliance Register ## General Information - **Responsible Person**: {answers['responsible_person']} - **Deployment Date**: {answers['deployment_date']} - **System Description**: {answers['ai_description']} ## Risk Category - **Type**: {answers['ai_type']} - **Risk Level**: {answers['risk_level']} - **Justification**: {answers['risk_justification']} ## Compliance Measures - **Data Evaluation**: {answers['data_evaluation']} - **Technical Docs**: {answers['technical_docs']} - **Human Oversight**: {answers['human_oversight']} - **Transparency Measures**: {answers['transparency_measures']} ## Audit & Follow-up - **Audit Frequency**: {answers['audit_frequency']} - **Compliance Contact**: {answers['compliance_contact']} --- Generated by AI Act Assistant. """ detected_lang = detect(filled) if filled.strip() else "en" csv_file = "ai_act_registers.csv" fieldnames = [key for key, _ in QUESTIONS] + ["timestamp"] row_data = {**answers, "timestamp": datetime.datetime.now().isoformat()} file_exists = os.path.isfile(csv_file) with open(csv_file, mode="a", newline="", encoding="utf-8") as f: writer = csv.DictWriter(f, fieldnames=fieldnames) if not file_exists: writer.writeheader() writer.writerow(row_data) pdf_path = export_text_to_pdf(filled, answers, language=detected_lang) return f"✅ Your PDF is ready for download.", {"done": True, "pdf": pdf_path}, pdf_path # === Gradio Interface === def launch_step_by_step_ui(): with gr.Blocks(title="AI Act Assistant", css="""footer, a[href*="gradio.app"], a[href*="huggingface.co"] { display: none !important; }""") as demo: gr.Markdown("### 🔒 GDPR Notice\nThis assistant does not store personal data. Use responsibly.") chatbot = gr.Chatbot(type="messages", value=[]) msg = gr.Textbox(label="Your answer") state = gr.State() file_output = gr.File(label="Download PDF", visible=True) restart = gr.Button("🔁 Restart") def chat_logic(user_msg, state): reply, updated_state, file_path = step_by_step_agent(user_msg, state) messages = [gr.ChatMessage(role="user", content=user_msg)] if isinstance(reply, str): messages.append(gr.ChatMessage(role="assistant", content=reply)) return messages, updated_state, file_path if file_path else None def reset(): first_q = QUESTIONS[0][1] return [gr.ChatMessage(role="assistant", content=f"👋 Let's get started.\n\n{first_q}")], {"step": 0, "answers": {}}, None msg.submit(chat_logic, [msg, state], [chatbot, state, file_output]) restart.click(reset, outputs=[chatbot, state, file_output]) demo.launch(show_api=False) def get_questions(): return QUESTIONS def run_tool(): return launch_step_by_step_ui()