data-science-document-ai 1.45.2__tar.gz → 1.59.0__tar.gz

{data_science_document_ai-1.45.2 → data_science_document_ai-1.59.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: data-science-document-ai
-Version: 1.45.2
+Version: 1.59.0
 Summary: "Document AI repo for data science"
 Author: Naomi Nguyen
 Author-email: naomi.nguyen@forto.com

{data_science_document_ai-1.45.2 → data_science_document_ai-1.59.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "data-science-document-ai"
-version = "1.45.2"
+version = "1.59.0"
 description = "\"Document AI repo for data science\""
 authors = ["Naomi Nguyen <naomi.nguyen@forto.com>", "Kumar Rajendrababu <kumar.rajendrababu@forto.com>", "Igor Tonko <igor.tonko@forto.com>", "Osman Demirel <osman.demirel@forto.com>"]
 packages = [

{data_science_document_ai-1.45.2 → data_science_document_ai-1.59.0}/src/constants.py

@@ -20,10 +20,11 @@ project_parameters = {
     # Fuzzy lookup
     "g_model_fuzzy_lookup_folder": "fuzzy_lookup",
     "item_code_lookup": "line_item_kvp_table.json",
+    "intermodal_partners": "intermodal_partners.json",
     "invoice_classification_lookup": "invoice_classification.json",
     "reverse_charge_sentence_lookup": "reverse_charge_sentences.json",
     # Fuzzy logic params
-    "fuzzy_threshold_item_code": 90,
+    "fuzzy_threshold_item_code": 92,
     "fuzzy_threshold_reverse_charge": 80,
     "fuzzy_threshold_invoice_classification": 70,
     # Chunking params
@@ -36,6 +37,8 @@ project_parameters = {
     # models metadata (confidence),
     "g_model_data_folder": "models",
     "local_model_data_folder": "data",
+    "if_use_docai": False,
+    "if_use_llm": True,  # Keep it always True
     "released_doc_types": {
         "bookingConfirmation",
         "packingList",
@@ -50,23 +53,6 @@ project_parameters = {
         "customsInvoice",
         "bundeskasse",
     },
-    "model_selector": {
-        "stable": {
-            "bookingConfirmation": 1,
-            "finalMbL": 0,
-            "draftMbl": 0,
-            "arrivalNotice": 0,
-            "shippingInstruction": 0,
-            "customsAssessment": 0,
-            "deliveryOrder": 0,
-            "partnerInvoice": 0,
-        },
-        "beta": {
-            "bookingConfirmation": 0,
-        },
-    },
-    # this is the model selector for the model to be used from the model_config.yaml
-    # file based on the environment, 0 mean the first model in the list
     # LLM model parameters
     "gemini_params": {
         "temperature": 0,
@@ -84,23 +70,15 @@ project_parameters = {
         "seed": 42,
         "model_id": "gemini-2.5-flash",
     },
-    # Key to combine the LLM results with the Doc Ai results
-    "key_to_combine": {
-        "bookingConfirmation": ["transportLegs"],
-        "finalMbL": ["containers"],
-        "draftMbl": ["containers"],
-        "customsAssessment": ["containers"],
-        "packingList": ["skuData"],
-        "commercialInvoice": ["skus"],
-        "shippingInstruction": ["containers"],
-        "partnerInvoice": ["lineItem"],
-        "customsInvoice": ["lineItem"],
-        "bundeskasse": ["lineItem"],
-    },
 }
 
 # Hardcoded rules for data points formatting that can't be based on label name alone
 formatting_rules = {
-    "bookingConfirmation": {"pickUpTerminal": "depot", "gateInTerminal": "terminal"},
+    "bookingConfirmation": {
+        "pickUpDepotCode": "depot",
+        "dropOffDepotCode": "depot",
+        "gateInTerminalCode": "terminal",
+        "pickUpTerminalCode": "terminal",
+    },
     "deliveryOrder": {"pickUpTerminal": "terminal", "EmptyContainerDepot": "depot"},
 }

data_science_document_ai-1.59.0/src/docai_processor_config.yaml

@@ -0,0 +1,9 @@
+models_project_id: "738250249861"
+model_config:
+  stable:
+    bookingConfirmation:
+      - id: "dc3e714cd168aeaa"
+        details:
+          display_name: "doc_cap_bookingConfirmation"
+          author: "reet.kanjilal@forto.com"
+          created_date: ""

{data_science_document_ai-1.45.2 → data_science_document_ai-1.59.0}/src/excel_processing.py

@@ -11,9 +11,8 @@ import asyncio
 import numpy as np
 import pandas as pd
 
-from src.llm import prompt_excel_extraction
 from src.prompts.prompt_library import prompt_library
-from src.utils import estimate_page_count, generate_schema_structure, get_excel_sheets
+from src.utils import estimate_page_count, get_excel_sheets
 
 
 async def extract_data_from_sheet(
@@ -29,11 +28,14 @@ async def extract_data_from_sheet(
     )
 
     # Prompt for the LLM JSON
-    prompt_docai = prompt_excel_extraction(worksheet)
+    prompt = prompt_library.library[doc_type]["other"]["prompt"]
+
+    # Join the worksheet content with the prompt
+    prompt = worksheet + "\n" + prompt
 
     try:
         result = await llm_client.get_unified_json_genai(
-            prompt_docai,
+            prompt,
             response_schema=response_schema,
             doc_type=doc_type,
         )
@@ -67,19 +69,7 @@ async def extract_data_from_excel(
 
     """
     # Generate the response structure
-    response_schema = (
-        prompt_library.library[input_doc_type]["other"]["placeholders"]
-        if input_doc_type
-        in [
-            "partnerInvoice",
-            "customsInvoice",
-            "bundeskasse",
-            "commercialInvoice",
-            "packingList",
-            "bookingConfirmation",
-        ]
-        else generate_schema_structure(params, input_doc_type)
-    )
+    response_schema = prompt_library.library[input_doc_type]["other"]["placeholders"]
 
     # Load the Excel file and get ONLY the "visible" sheet names
     sheets, workbook = get_excel_sheets(file_content, mime_type)

{data_science_document_ai-1.45.2 → data_science_document_ai-1.59.0}/src/llm.py

@@ -201,33 +201,4 @@ class LlmClient:
         return response
 
 
-def prompt_excel_extraction(excel_structured_text):
-    """Write a prompt to extract data from Excel files.
-
-    Args:
-        excel_structured_text (str): The structured text of the Excel file.
-
-    Returns:
-        prompt str: The prompt for common json.
-    """
-    prompt = f"""{excel_structured_text}
-
-    Task: Fill in the following dictionary from the information in the given in the above excel data.
-
-    Instructions:
-    - Do not change the keys of the following dictionary.
-    - The values should be filled in as per the schema provided below.
-    - If an entity contains a 'display_name', consider its properties as child data points in the below format.
-    {{'data-field': {{
-        'child-data-field': 'type -occurrence_type- description',
-          }}
-    }}
-    - The entity with 'display_name' can be extracted multiple times. Please pay attention to the occurrence_type.
-    - Ensure the schema reflects the hierarchical relationship.
-    - Use the data field description to understand the context of the data.
-
-    """
-    return prompt
-
-
 # pylint: enable=all

{data_science_document_ai-1.45.2 → data_science_document_ai-1.59.0}/src/pdf_processing.py

@@ -32,7 +32,6 @@ from src.postprocessing.postprocess_partner_invoice import (
 from src.prompts.prompt_library import prompt_library
 from src.utils import (
     extract_top_pages,
-    generate_schema_structure,
     get_pdf_page_count,
     get_processor_name,
     run_background_tasks,
@@ -202,21 +201,6 @@ async def process_file_w_llm(params, file_content, input_doc_type, llm_client):
     number_of_pages = get_pdf_page_count(file_content)
     logger.info(f"processing {input_doc_type} with {number_of_pages} pages...")
 
-    # get the schema placeholder from the Doc AI and generate the response structure
-    response_schema = (
-        prompt_library.library[input_doc_type]["other"]["placeholders"]
-        if input_doc_type
-        in [
-            "partnerInvoice",
-            "customsInvoice",
-            "bundeskasse",
-            "commercialInvoice",
-            "packingList",
-            "bookingConfirmation",
-        ]
-        else generate_schema_structure(params, input_doc_type)
-    )
-
     carrier = "other"
     carrier_schema = (
         prompt_library.library.get("preprocessing", {})
@@ -253,6 +237,9 @@ async def process_file_w_llm(params, file_content, input_doc_type, llm_client):
     # get the related prompt from predefined prompt library
     prompt = prompt_library.library[input_doc_type][carrier]["prompt"]
 
+    # get the schema placeholder
+    response_schema = prompt_library.library[input_doc_type][carrier]["placeholders"]
+
     # Add page-number extraction for moderately large docs
     use_chunking = number_of_pages >= params["chunk_after"]
 
@@ -270,7 +257,11 @@ async def process_file_w_llm(params, file_content, input_doc_type, llm_client):
     ):
         tasks.append(
             process_chunk_with_retry(
-                chunk, prompt, response_schema, llm_client, input_doc_type
+                chunk,
+                prompt,
+                response_schema,
+                llm_client,
+                input_doc_type,
             )
         )
 
@@ -362,8 +353,7 @@ async def extract_data_from_pdf_w_llm(params, input_doc_type, file_content, llm_
     # Add currency from the amount field
     if input_doc_type in ["commercialInvoice"]:
         result = postprocessing_commercial_invoice(result, params, input_doc_type)
-    elif input_doc_type == "bookingConfirmation":
-        result = postprocess_booking_confirmation(result)
+
     return result, llm_client.model_id
 
 
@@ -382,13 +372,14 @@ def combine_llm_results_w_doc_ai(
     Returns:
         combined result
     """
-    result = doc_ai.copy()
-    llm = remove_none_values(llm)
-    if not llm:
+    result = remove_none_values(llm)
+
+    docAi = doc_ai.copy()
+    if not docAi:
         return result
 
     # Merge top-level keys
-    result.update({k: v for k, v in llm.items() if k not in result})
+    result.update({k: v for k, v in docAi.items() if k not in result})
 
     if (
         input_doc_type
@@ -396,28 +387,28 @@ def combine_llm_results_w_doc_ai(
         and keys_to_combine
     ):
         result.update(
-            {key: llm.get(key) for key in keys_to_combine if key in llm.keys()}
+            {key: docAi.get(key) for key in keys_to_combine if key in docAi.keys()}
         )
         return result
 
     # Handle specific key-based merging logic for multiple keys
     if keys_to_combine:
         for key in keys_to_combine:
-            if key in llm.keys():
+            if key in docAi.keys():
                 # Merge the list of dictionaries
-                # If the length of the LLM list is less than the Doc AI result, replace with the LLM list
-                if len(llm[key]) < len(result[key]):
-                    result[key] = llm[key]
+                # If the length of the docAi list is less than the LLM result, replace with the docAi list
+                if len(docAi[key]) < len(result[key]):
+                    result[key] = docAi[key]
                 else:
-                    # If the length of the LLM list is greater than or equal to the Doc AI result,
+                    # If the length of the docAi list is greater than or equal to the LLM result,
                     # add & merge the dictionaries
-                    if isinstance(llm[key], list):
-                        for i in range(len(llm[key])):
+                    if isinstance(docAi[key], list):
+                        for i in range(len(docAi[key])):
                             if i == len(result[key]):
-                                result[key].append(llm[key][i])
+                                result[key].append(docAi[key][i])
                             else:
-                                for sub_key in llm[key][i].keys():
-                                    result[key][i][sub_key] = llm[key][i][sub_key]
+                                for sub_key in docAi[key][i].keys():
+                                    result[key][i][sub_key] = docAi[key][i][sub_key]
     return result
 
 
@@ -511,13 +502,15 @@ async def data_extraction_manual_flow(
     page_count = None
     # Validate the file type
     if mime_type == "application/pdf":
+        if_use_docai = params["if_use_docai"]
+
         # Enable Doc Ai only for certain document types.
-        if_use_docai = (
-            True if meta.documentTypeCode in params["model_config"]["stable"] else False
-        )
-        if_use_llm = (
-            True if meta.documentTypeCode in params["key_to_combine"].keys() else False
-        )
+        if params["if_use_docai"]:
+            if_use_docai = (
+                True
+                if meta.documentTypeCode in params["model_config"]["stable"]
+                else False
+            )
 
         (
             extracted_data,
@@ -529,7 +522,7 @@ async def data_extraction_manual_flow(
             meta.documentTypeCode,
             processor_client,
             if_use_docai=if_use_docai,
-            if_use_llm=if_use_llm,
+            if_use_llm=params["if_use_llm"],
             llm_client=llm_client,
             isBetaTest=False,
         )

{data_science_document_ai-1.45.2 → data_science_document_ai-1.59.0}/src/postprocessing/postprocess_partner_invoice.py

@@ -1,4 +1,6 @@
 """This module contains the postprocessing functions for the partner invoice."""
+from collections import defaultdict
+
 from rapidfuzz import fuzz, process
 
 from src.io import logger
@@ -103,9 +105,18 @@ def post_process_bundeskasse(aggregated_data):
             )
 
         # Check if the deferredDutyPayer is forto
-        deferredDutyPayer = line_item.get("deferredDutyPayer", {})
-        lower = deferredDutyPayer.get("documentValue", "").lower()
-        if any(key in lower for key in ["de789147263644738", "forto"]):
+        KEYWORDS = {"de789147263644738", "forto", "009812"}
+
+        def is_forto_recipient(line_item: dict) -> bool:
+            values_to_check = [
+                line_item.get("deferredDutyPayer", {}).get("documentValue", ""),
+                line_item.get("vatId", {}).get("documentValue", ""),
+            ]
+
+            combined = " ".join(values_to_check).lower()
+            return any(keyword in combined for keyword in KEYWORDS)
+
+        if is_forto_recipient(line_item):
             is_recipient_forto = True
 
     update_recipient_and_vendor(aggregated_data, is_recipient_forto)
@@ -134,6 +145,20 @@ def update_recipient_and_vendor(aggregated_data, is_recipient_forto):
     ] = "Dasbachstraße 15, 54292 Trier, Germany"
 
 
+def select_unique_bank_account(bank_account):
+    # Select the unique bank account if multiple are present
+    if isinstance(bank_account, list) and bank_account:
+        best = defaultdict(lambda: None)
+
+        for item in bank_account:
+            dv = item["documentValue"]
+            if best[dv] is None or item["page"] < best[dv]["page"]:
+                best[dv] = item
+
+        unique = list(best.values())
+        return unique
+
+
 async def process_partner_invoice(params, aggregated_data, document_type_code):
     """Process the partner invoice data."""
     # Post process bundeskasse invoices
@@ -141,6 +166,11 @@ async def process_partner_invoice(params, aggregated_data, document_type_code):
         post_process_bundeskasse(aggregated_data)
         return
 
+    if "bankAccount" in aggregated_data:
+        aggregated_data["bankAccount"] = select_unique_bank_account(
+            aggregated_data["bankAccount"]
+        )
+
     line_items = aggregated_data.get("lineItem", [])
     # Add debug logging
     logger.info(f"Processing partnerInvoice with {len(line_items)} line items")
@@ -158,15 +188,20 @@ async def process_partner_invoice(params, aggregated_data, document_type_code):
         reverse_charge_info["formattedValue"] = reverse_charge_value
         reverse_charge = aggregated_data.pop("reverseChargeSentence", None)
 
+    # Partner Name
+    partner_name = aggregated_data.get("vendorName", {}).get("documentValue", None)
+
     # Process everything in one go
-    processed_items = await process_line_items_batch(params, line_items, reverse_charge)
+    processed_items = await process_line_items_batch(
+        params, line_items, reverse_charge, partner_name
+    )
 
     # Update your main data structure
     aggregated_data["lineItem"] = processed_items
 
 
 async def process_line_items_batch(
-    params: dict, line_items: list[dict], reverse_charge=None
+    params: dict, line_items: list[dict], reverse_charge=None, partner_name=None
 ):
     """
     Processes all line items efficiently using a "Split-Apply-Combine" strategy.
@@ -204,23 +239,12 @@ async def process_line_items_batch(
 
     # Batch API Call for Embedding lookups
     if pending_line_items:
-        values_to_fetch = list(set(pending_line_items.values()))
-        logger.info(f"Mapping {len(values_to_fetch)} line items from Embedding API...")
-
-        # Await the batch response {"desc1": "code1", "desc2": "code2"}
-        api_results = await get_tms_mappings(
-            input_list=values_to_fetch, embedding_type="line_items"
-        )
+        code_map = await fetch_line_item_codes(pending_line_items, partner_name, params)
 
-        # Merge API results back into original list
         for index, desc in pending_line_items.items():
-            # Get result from API response, or None if API failed for that item
-            forto_code = api_results.get(desc)
-
-            # Update the original item
             line_items[index]["itemCode"] = {
                 "documentValue": desc,
-                "formattedValue": forto_code,  # Might be None if API failed
+                "formattedValue": code_map.get(desc),
                 "page": line_items[index]["lineItemDescription"].get("page"),
             }
 
@@ -229,8 +253,12 @@ async def process_line_items_batch(
         [
             item.update({"reverseChargeSentence": reverse_charge})
             for item in line_items
-            if item["itemCode"]["formattedValue"] != "CDU"
+            if (
+                (item.get("itemCode") and item["itemCode"]["formattedValue"] != "CDU")
+                or not item.get("itemCode")
+            )
         ]
+
     return line_items
 
 
@@ -272,11 +300,14 @@ def if_reverse_charge_sentence(sentence: str, params):
         return False
 
     # Check if the sentence is similar to any of the reverse charge sentences
-    _, is_reverse_charge = get_fuzzy_match_score(
-        sentence, reverse_charge_sentences, threshold
+    match, _ = get_fuzzy_match_score(
+        sentence, list(reverse_charge_sentences.keys()), threshold
     )
 
-    return is_reverse_charge
+    if match:
+        return reverse_charge_sentences[match]
+
+    return False
 
 
 def find_matching_lineitem(new_lineitem: str, kvp_dict: dict, threshold=90):
@@ -307,12 +338,13 @@ def find_matching_lineitem(new_lineitem: str, kvp_dict: dict, threshold=90):
     return None
 
 
-async def associate_forto_item_code(line_item_data, params):
+async def associate_forto_item_code(line_item_data, params, partner_name=None):
     """
     Associates Forto item codes to a list of line item descriptions.
     Args:
         line_item_data (dict): A dictionary where keys are original descriptions and values are cleaned descriptions.
         params (dict): Parameters containing lookup data and thresholds.
+        partner_name (str, optional): The name of the partner for context in matching. Defaults to None.
 
     Returns:
         list: A list of dictionaries with 'description' and 'itemCode' keys.
@@ -334,14 +366,51 @@ async def associate_forto_item_code(line_item_data, params):
 
     # Batch API Call for Embedding lookups
     if pending_line_items:
-        api_results = await get_tms_mappings(
-            input_list=list(pending_line_items.values()),
-            embedding_type="line_items",
-        )
+        code_map = await fetch_line_item_codes(pending_line_items, partner_name, params)
 
-        # Merge API results back into original list
         for desc, f_desc in pending_line_items.items():
-            code = api_results.get(f_desc)
-            result.append({"description": desc, "itemCode": code})
+            result.append(
+                {
+                    "description": desc,
+                    "itemCode": code_map.get(f_desc),
+                }
+            )
+
+    return result
+
+
+async def fetch_line_item_codes(
+    pending_line_items: dict,
+    partner_name: str | None,
+    params: dict,
+):
+    """Returns: {original_description: mapped_code_or_None}"""
+    t_mode = (
+        find_matching_lineitem(
+            partner_name.upper(),
+            params["lookup_data"]["intermodal_partners"],
+            threshold=87,
+        )
+        if partner_name
+        else None
+    )
 
+    unique_descs = list(set(pending_line_items.values()))
+    logger.info(f"Mapping {len(unique_descs)} line items from Embedding API...")
+
+    # Build API input map
+    api_input_map = {
+        desc: f"{t_mode} - {desc}" if t_mode else desc for desc in unique_descs
+    }
+
+    api_results = await get_tms_mappings(
+        input_list=list(api_input_map.values()),
+        embedding_type="line_items",
+    )
+
+    # Normalize response back to original descriptions
+    result = {
+        original_desc: api_results.get(api_desc)
+        for original_desc, api_desc in api_input_map.items()
+    }
     return result

data_science_document_ai-1.59.0/src/prompts/library/arrivalNotice/other/placeholders.json

@@ -0,0 +1,70 @@
+{
+  "type": "OBJECT",
+  "properties": {
+      "bookingNumber": {
+        "type": "STRING",
+        "nullable": true,
+        "description": "The booking number associated with the Arrival Notice document. They are often referred to as 'Booking Number', 'Booking No.', 'Booking Ref.', 'Booking Reference', 'Booking ID', 'carrier's reference' or 'Order Ref'."
+      },
+      "destinationTerminal": {
+        "type": "STRING",
+        "nullable": true,
+        "description": "The terminal at the destination port where the container will be delivered."
+      },
+      "eta": {
+        "type": "STRING",
+        "nullable": true,
+        "description": "Estimated Time of Arrival (ETA) is the expected date when the shipment will arrive at its destination."
+      },
+      "mblNumber": {
+        "type": "STRING",
+        "nullable": true,
+        "description": "Bill of Lading number (B/L NO.), a document issued by the carrier."
+      },
+      "portOfDischarge": {
+        "type": "STRING",
+        "nullable": true,
+        "description": "The port where the goods are discharged from the vessel. This is the destination port for the shipment."
+      },
+      "vesselName": {
+        "type": "STRING",
+        "nullable": true,
+        "description": "The name of the vessel carrying the shipment."
+      },
+      "containers": {
+          "type": "ARRAY",
+          "items": {
+              "type": "OBJECT",
+              "properties": {
+                  "containerNumber": {
+                    "type": "STRING",
+                    "nullable": true,
+                    "description": "The unique identifier for each container. It always starts with 4 capital letters and followed by 7 digits. Example: TEMU7972458."
+                  },
+                  "containerType": {
+                    "type": "STRING",
+                    "nullable": true,
+                    "description": "The size of the container associated with the containerNumber, such as 20ft, 40ft, 40HC, 20DC etc."
+                  },
+                  "grossWeight": {
+                    "type": "STRING",
+                    "nullable": true,
+                    "description": "The gross weight of the container. Usually mentioned as G.W or GW or Gross Weight, etc.."
+                  },
+                  "measurements": {
+                    "type": "STRING",
+                    "nullable": true,
+                    "description": "The volume of the container. Usually, it is measured in 'Cubic Meter (cbm)' or dimensions. But volume in 'cbm' is preferred."
+                  },
+                  "sealNumber": {
+                    "type": "STRING",
+                    "nullable": true,
+                    "description": "The seal number associated with the container Number. But it is not same as the container number."
+                  }
+              },
+              "required": ["containerNumber", "containerType", "grossWeight"]
+          }
+      }
+},
+  "required": ["bookingNumber", "destinationTerminal", "eta", "portOfDischarge", "vesselName", "containers"]
+}

data_science_document_ai-1.59.0/src/prompts/library/arrivalNotice/other/prompt.txt

@@ -0,0 +1,40 @@
+<PERSONA> You are an efficient document entity data extraction specialist working for a Freight Forwarding company. <PERSONA>
+
+<TASK> Your task is to extract data from Arrival Notice documents as per the given response schema structure. <TASK>
+
+<CONTEXT>
+The Freight Forwarding company receives Arrival Notice from shipping lines.
+These documents contain various details related to arrival of a shipment to the port of destination such as container numbers, estimated time of arrival, vessel details and containers information.
+They may be written in different languages such as English, German, Italian and can appear in a variety of formats and layouts.
+Your role is to accurately extract specific entities from these Arrival Notices to support efficient processing and accurate record-keeping.
+<CONTEXT>
+
+<INSTRUCTIONS>
+- Populate fields as defined in the response schema.
+- Multiple Containers entries may exist, capture all instances under "containers".
+- Use the data field description to understand the context of the data.
+
+- bookingNumbers:
+    - Booking numbers are unique identifiers for shipments. They are often referred to as "Booking Number", "Booking No.", "Booking Ref.", "Booking Reference", "Booking ID", "SACO-Pos.", "Order Ref", "Unsere Referenz", or "Unsere Position"
+    - If there is a unique_id that starts with "S" followed by 6 or 8 digits, it is a shipmentID, not a bookingNumber.
+
+- destinationTerminal:
+    - Destination Terminal can also be referred to as "Destination Termina;", "Pickup Location", "Delivery Location", "Delivery Terminal", "Empfangsort", "Entladeort", or "Abladestelle".
+
+- mblNumbers:
+    - Commonly known as "Bill of Lading Number", "BILL OF LADING NO.", "BL Number", "BL No.", "B/L No.", "BL-Nr.", "B/L", "HBL No.", or "M-AWB Nummer".
+    - Bill of Lading Number is known as mblNumber. Not a shipmentID even if it starts with "S".
+    - mblNumber from Hapag-Lloyd always starts with HLC.... (e.g., "HLCUTS12303AWNT3) and named as SEA WAYBILL or "SWB-NR.
+
+- eta:
+    - Estimated Time of Arrival (ETA) is the expected date and time when the shipment will arrive at the destination port.
+    - It can be referred to as "ETA", "Estimated Arrival", "Voraussichtliche Ankunft", "Ankunftszeit", "Arrivo", "Due to arrive at Terminal"
+
+- vesselName:
+    - Vessel Name is the name of the ship carrying the cargo. It can be referred to as "Vessel", "Ship Name", "Schiff", "Schiffsname", "Nave", or "Vessel/Flight No.".
+
+- containers: Details of each container on the arrival notice. Make sure to extract each container information separately.
+    - containerNumber: Container Number consists of 4 capital letters followed by 7 digits (e.g., TEMU7972458, CAIU 7222892).
+    - sealNumber: Seal numbers are unique identifiers for shipping seals. They are usually mentioned as seal numbers in the document but they are definitely not container numbers.
+
+<INSTRUCTIONS>