Spaces:
Running
Running
| import streamlit as st | |
| import requests | |
| # ----------------- Streamlit Setup ----------------- # | |
| st.set_page_config(page_title="Lease Agreement Extractor (OCR + AI)", layout="centered") | |
| st.title("📄 Lease Agreement Extractor (OCR + AI)") | |
| # ----------------- Assistant & API Config ----------------- # | |
| OPENAI_ASSISTANT_ID = "asst_xBnNfiyWmVa4iF3CgXwJnmBt" # Replace with your assistant ID | |
| OPENAI_API_KEY = "YOUR_OPENAI_API_KEY" # Replace with your OpenAI key | |
| HEADERS = { | |
| "Authorization": f"Bearer {OPENAI_API_KEY}", | |
| "Content-Type": "application/json" | |
| } | |
| # ----------------- Upload Interface ----------------- # | |
| uploaded_file = st.file_uploader("Upload a lease agreement (PDF, PNG, JPG)", type=["pdf", "png", "jpg", "jpeg"]) | |
| def upload_file_to_openai(file_bytes, filename): | |
| files = { | |
| "file": (filename, file_bytes, "application/octet-stream") | |
| } | |
| response = requests.post("https://api.openai.com/v1/files", headers={"Authorization": f"Bearer {OPENAI_API_KEY}"}, files=files) | |
| return response.json().get("id") | |
| def create_thread_and_run(file_id): | |
| # Create thread | |
| thread_res = requests.post("https://api.openai.com/v1/threads", headers=HEADERS, json={}) | |
| thread_data = thread_res.json() | |
| thread_id = thread_data.get("id") | |
| # Run assistant with structured lease extraction prompt | |
| run_payload = { | |
| "assistant_id": OPENAI_ASSISTANT_ID, | |
| "instructions": """ | |
| You are an AI assistant designed to extract structured lease data from any uploaded agreement, even if it is scanned. Your responsibilities are: | |
| - Accept PDF or image-based lease documents. | |
| - Use OCR (Optical Character Recognition) if the document is non-selectable or scanned. | |
| - Parse and extract key structured data fields from lease agreements. | |
| - Output results in JSON format following a fixed schema. | |
| OCR & Extraction Rules: | |
| 1. If text cannot be extracted normally, perform OCR using image-to-text on each page. | |
| 2. Preserve paragraph structure, detect headers, signature blocks, tables, and dates. | |
| Expected JSON Output: | |
| { | |
| "document_title": "Lease Agreement Title or File Name", | |
| "parties": { | |
| "lessor": "Name of Lessor", | |
| "lessee": "Name of Lessee" | |
| }, | |
| "property_description": "Detailed description of leased property", | |
| "term": { | |
| "start_date": "YYYY-MM-DD", | |
| "end_date": "YYYY-MM-DD", | |
| "renewal_options": "Yes/No or Clause Text" | |
| }, | |
| "financials": { | |
| "monthly_rent": "Amount", | |
| "deposit": "Amount", | |
| "escalation_clause": "Text if present" | |
| }, | |
| "obligations": { | |
| "maintenance": "Lessor/Lessee", | |
| "insurance": "Lessor/Lessee", | |
| "subletting": "Allowed/Not allowed" | |
| }, | |
| "signatures": [ | |
| { | |
| "party": "Party Name", | |
| "signed_by": "Full Name", | |
| "date": "YYYY-MM-DD", | |
| "position": "Title if listed" | |
| } | |
| ], | |
| "raw_text_per_page": { | |
| "page_1": "Full OCR text of page 1", | |
| "page_2": "Full OCR text of page 2" | |
| } | |
| } | |
| 📌 Behaviors: | |
| - Automatically detect if OCR is needed. | |
| - Normalize dates to YYYY-MM-DD. | |
| - Use "Not Found" or null where info is missing. | |
| - Return JSON only. | |
| - If OCR text is low-confidence, include: "low_confidence": true | |
| """, | |
| "file_ids": [file_id] | |
| } | |
| run_res = requests.post(f"https://api.openai.com/v1/threads/{thread_id}/runs", headers=HEADERS, json=run_payload) | |
| run_data = run_res.json() | |
| run_id = run_data.get("id") | |
| return thread_id, run_id | |
| # ----------------- Main UI Flow ----------------- # | |
| if uploaded_file is not None: | |
| st.success("📄 File uploaded successfully.") | |
| if st.button("🚀 Run Lease Extraction"): | |
| with st.spinner("Uploading and invoking assistant..."): | |
| file_id = upload_file_to_openai(uploaded_file.getvalue(), uploaded_file.name) | |
| if file_id: | |
| thread_id, run_id = create_thread_and_run(file_id) | |
| st.success("✅ Assistant run started!") | |
| st.code(f"Thread ID: {thread_id}", language="text") | |
| st.code(f"Run ID: {run_id}", language="text") | |
| st.markdown("📌 You can retrieve results via OpenAI API using these IDs.") | |
| else: | |
| st.error("❌ File upload to OpenAI failed.") | |
| else: | |
| st.info("Please upload a lease agreement file to begin.") | |