diff --git a/README.md b/README.md
index 2366626..c14cd42 100644
--- a/README.md
+++ b/README.md
@@ -1,4 +1,11 @@
-# aicup-rag
+# AI CUP 2024 玉山人工智慧公開挑戰賽－RAG與LLM在金融問答的應用
+
+## Result
+
+- Total: 38 / 487 Teams
+   - Leaderboard: 38 / 222
+
+![AI Cup Result](img/aicup_result.png)
 
 ## Development Mode
 To set up the development environment, follow these steps:
diff --git a/data/conbine_result.py b/data/conbine_result.py
index 0f72467..a021ced 100644
--- a/data/conbine_result.py
+++ b/data/conbine_result.py
@@ -1,31 +1,33 @@
 import json
 
 # 載入 aicup_noocr.json 和 aicup_ref.json
-with open('data/aicup_noocr.json', 'r', encoding='utf-8') as file:
+with open('data/aicup_noocr.json', encoding='utf-8') as file:
     noocr_data = json.load(file)
 
-with open('data/aicup_ref.json', 'r', encoding='utf-8') as file:
+with open('data/aicup_ref.json', encoding='utf-8') as file:
     ref_data = json.load(file)
 
 # 建立 ref_data 的 dictionary，並檢查 content 是否為字串，再去除空格
 ref_dict = {
-    (item["category"], item["pid"]): ''.join(item["content"].split()) if isinstance(item["content"], str) else item["content"]
+    (item['category'], item['pid']): ''.join(item['content'].split())
+    if isinstance(item['content'], str)
+    else item['content']
     for item in ref_data
 }
 
 # 更新 noocr_data 中空的 content
 for item in noocr_data:
-    category = item["category"]
-    pid = item["pid"]
-    content = item["content"]
+    category = item['category']
+    pid = item['pid']
+    content = item['content']
 
     # 如果 content 是 string 並且為空，則從 ref_data 裡填入去掉空格的 content
-    if isinstance(content, str) and content == "":
+    if isinstance(content, str) and content == '':
         if (category, pid) in ref_dict:
-            item["content"] = ref_dict[(category, pid)]
+            item['content'] = ref_dict[(category, pid)]
 
 # 將結果寫入 aicup_noocr_sec.json
 with open('data/aicup_noocr_sec.json', 'w', encoding='utf-8') as file:
     json.dump(noocr_data, file, ensure_ascii=False, indent=4)
 
-print("已完成比對並生成 aicup_noocr_sec.json，並移除轉入的 content 中的空格（如果 content 是字串）")
+print('已完成比對並生成 aicup_noocr_sec.json，並移除轉入的 content 中的空格（如果 content 是字串）')
diff --git a/data/read_data_noocr.py b/data/read_data_noocr.py
index 7b254d9..cbc744e 100644
--- a/data/read_data_noocr.py
+++ b/data/read_data_noocr.py
@@ -1,8 +1,10 @@
-import os
 import json
+import os
+
 import pdfplumber
 from tqdm import tqdm
 
+
 # 讀取單個PDF文件並返回其文本內容
 def read_pdf(pdf_loc):
     pdf = pdfplumber.open(pdf_loc)
@@ -14,6 +16,7 @@ def read_pdf(pdf_loc):
     pdf.close()
     return pdf_text
 
+
 # 從指定資料夾載入PDF文件，並根據資料夾名稱設定category
 def load_data_by_category(source_path, category):
     pdf_files = [f for f in os.listdir(source_path) if f.endswith('.pdf')]
@@ -21,17 +24,18 @@ def load_data_by_category(source_path, category):
     for file in tqdm(pdf_files):
         pid = file.replace('.pdf', '')  # 擷取檔案名稱作為pid
         content = read_pdf(os.path.join(source_path, file))  # 讀取PDF內文
-        data.append({"category": category, "pid": pid, "content": content})
+        data.append({'category': category, 'pid': pid, 'content': content})
     return data
 
+
 # 主程式
 def generate_json(output_path):
     all_data = []
-    
+
     # 載入不同類別的PDF資料
     source_paths = {
-        "finance": "reference/finance",     # finance 資料夾的路徑
-        "insurance": "reference/insurance"  # insurance 資料夾的路徑
+        'finance': 'reference/finance',  # finance 資料夾的路徑
+        'insurance': 'reference/insurance',  # insurance 資料夾的路徑
     }
 
     # 遍歷每個類別的資料夾並載入資料
@@ -43,6 +47,7 @@ def generate_json(output_path):
     with open(output_path, 'w', encoding='utf8') as f:
         json.dump(all_data, f, ensure_ascii=False, indent=4)
 
+
 # 設定輸出路徑
 output_path = 'data/aicup_noocr.json'
 generate_json(output_path)
diff --git a/img/aicup_result.png b/img/aicup_result.png
new file mode 100644
index 0000000..b4a3e7f
Binary files /dev/null and b/img/aicup_result.png differ
diff --git a/src/db_insert.py b/src/db_insert.py
index ca5d6d0..76ec8a0 100644
--- a/src/db_insert.py
+++ b/src/db_insert.py
@@ -1,8 +1,9 @@
-import time
 import json
-from langchain.text_splitter import RecursiveCharacterTextSplitter
+import time
+
 import utils.config_log as config_log
 import weaviate
+from langchain.text_splitter import RecursiveCharacterTextSplitter
 
 config, logger, CONFIG_PATH = config_log.setup_config_and_logging()
 config.read(CONFIG_PATH)
@@ -13,6 +14,7 @@
 # Token limit for OpenAI model
 TOKEN_LIMIT = 8192
 
+
 class WeaviateManager:
     def __init__(self, classnm):
         self.url = wea_url
@@ -28,7 +30,11 @@ def check_class_exist(self):
             'class': self.classnm,
             'properties': [
                 {'name': 'pid', 'dataType': ['text']},
-                {'name': 'content', 'dataType': ['text'], "tokenization": "gse"}, # `gse` implements the "Jieba" algorithm, which is a popular Chinese text segmentation algorithm.
+                {
+                    'name': 'content',
+                    'dataType': ['text'],
+                    'tokenization': 'gse',
+                },  # `gse` implements the "Jieba" algorithm, which is a popular Chinese text segmentation algorithm.
             ],
             'vectorizer': 'text2vec-openai',
             'moduleConfig': {
@@ -51,19 +57,19 @@ def insert_data(self, pid, content):
                 error_msg = str(e)
                 # 檢查是否是因為 token 長度過長
                 if 'maximum context length' in error_msg:
-                    print(f"Content too long for pid: {pid}. Splitting content.")
-                    return "TOO_LONG"  # 特殊回傳值表達需要分割
+                    print(f'Content too long for pid: {pid}. Splitting content.')
+                    return 'TOO_LONG'  # 特殊回傳值表達需要分割
                 elif '429' in error_msg:
                     print(f'Rate limit exceeded, retrying in 5 seconds... (Attempt {attempt + 1}/{max_retries})')
                     time.sleep(5)
                 else:
-                    print(f"Unexpected Error for pid: {pid} - {error_msg}")
+                    print(f'Unexpected Error for pid: {pid} - {error_msg}')
                     return False
             except Exception as e:
                 print(f'Error inserting data for pid: {pid}, category: {self.classnm} - {str(e)}')
                 return False
         # 超過最大重試次數
-        print(f"Failed to insert data for pid: {pid} after {max_retries} attempts.")
+        print(f'Failed to insert data for pid: {pid} after {max_retries} attempts.')
         return False
 
     def split_and_insert(self, pid, content, category):
@@ -73,10 +79,10 @@ def split_and_insert(self, pid, content, category):
 
         # 逐段插入分割後的文本，保持相同的 pid 和 category
         for idx, part in enumerate(split_content):
-            print(f"Inserting split content part {idx + 1} for pid: {pid}")
+            print(f'Inserting split content part {idx + 1} for pid: {pid}')
             success = self.insert_data(pid, part)
             if not success:
-                failed_records.append({"pid": pid, "category": category})
+                failed_records.append({'pid': pid, 'category': category})
 
 
 if __name__ == '__main__':
@@ -90,27 +96,27 @@ def split_and_insert(self, pid, content, category):
         pid = item['pid']
         content = item['content']
 
-        if category == "faq":
-            classnm = "faqdev"
+        if category == 'faq':
+            classnm = 'faqdev'
             content_str = json.dumps(content, ensure_ascii=False, indent=4)
-        elif category == "insurance":
-            classnm = "insurancedev"
+        elif category == 'insurance':
+            classnm = 'insurancedev'
             content_str = content
-        elif category == "finance":
-            classnm = "financedev"
+        elif category == 'finance':
+            classnm = 'financedev'
             content_str = json.dumps(content, ensure_ascii=False, indent=4) if isinstance(content, dict) else content
         else:
-            print("Unknown category, skipping item.")
+            print('Unknown category, skipping item.')
             continue
 
         manager = WeaviateManager(classnm)
         result = manager.insert_data(pid, content_str)
 
         # 如果內容過長需要切割
-        if result == "TOO_LONG":
+        if result == 'TOO_LONG':
             manager.split_and_insert(pid, content_str, category)
         elif not result:  # 如果失敗且非長度問題
-            failed_records.append({"pid": pid, "category": category})
+            failed_records.append({'pid': pid, 'category': category})
 
     # 將失敗的資料寫入 JSON 檔案
     if failed_records:
@@ -118,4 +124,4 @@ def split_and_insert(self, pid, content, category):
             json.dump(failed_records, f, ensure_ascii=False, indent=4)
         print("Failed records have been written to 'failed_imports.json'")
     else:
-        print("All records imported successfully.")
+        print('All records imported successfully.')
diff --git a/src/flask_app.py b/src/flask_app.py
index 2e4b0f5..84058ea 100644
--- a/src/flask_app.py
+++ b/src/flask_app.py
@@ -5,6 +5,7 @@
 from flask_limiter import Limiter
 from flask_limiter.util import get_remote_address
 from flask_restx import Api, Resource, fields
+
 # from utils.weaviate_op import search_do
 from utils.weaviatexreranker import search_do
 from werkzeug.security import check_password_hash, generate_password_hash
@@ -45,7 +46,7 @@ def verify_password(username, password):
         'qid': fields.Integer(required=True, description='qid of the question'),
         'source': fields.List(fields.Integer, required=True, description='source of the question'),
         'query': fields.String(required=True, description='The message to the chatbot'),
-        'category': fields.String(required=True, description='The category of the question')
+        'category': fields.String(required=True, description='The category of the question'),
     },
 )
 
@@ -77,7 +78,7 @@ def post(self):
         # "query": "匯款銀行及中間行所收取之相關費用由誰負擔?",
         # "category": "insurance"
         # },
-        
+
         alpha = 0.5
 
         if not question:
@@ -87,10 +88,7 @@ def post(self):
         else:
             try:
                 response = search_do(question, category, source, alpha)
-                response = {
-                    'qid': qid,
-                    'retrieve': int(response)
-                }
+                response = {'qid': qid, 'retrieve': int(response)}
 
                 response = jsonify(response)
 
diff --git a/src/tools/automate.py b/src/tools/automate.py
index b2b8a04..b61aa2b 100644
--- a/src/tools/automate.py
+++ b/src/tools/automate.py
@@ -1,51 +1,49 @@
-import requests
 import json
 import time  # Import time module for timing
 
+import requests
+
 # Load questions from the JSON file
-with open('data/questions_example.json', 'r', encoding='utf-8') as file:
+with open('data/questions_example.json', encoding='utf-8') as file:
     questions = json.load(file)['questions']
 
-output_data = {"answers": []}  # Initialize output format with "answers" array
+output_data = {'answers': []}  # Initialize output format with "answers" array
 
-url = "http://127.0.0.1:5000/api/chat"
+url = 'http://127.0.0.1:5000/api/chat'
 
 total_start_time = time.time()  # Start timing for the entire process
 
 for question in questions:
     question_start_time = time.time()  # Start timing for each question
-    
+
     # Send POST request
     response = requests.post(url, json=question)
 
     if response.status_code == 200:
         response_json = response.json()
-        
+
         # Extract qid and retrieve from the API response
-        qid = question.get("qid")  # Assuming each question has a unique "qid" field
-        retrieve = response_json.get("retrieve")
-        
+        qid = question.get('qid')  # Assuming each question has a unique "qid" field
+        retrieve = response_json.get('retrieve')
+
         # Append formatted result to the answers array
-        output_data["answers"].append({
-            "qid": qid,
-            "retrieve": retrieve
-        })
-        print("成功取得 JSON:", response_json)
+        output_data['answers'].append({'qid': qid, 'retrieve': retrieve})
+        print('成功取得 JSON:', response_json)
     else:
-        print("請求失敗，狀態碼:", response.status_code)
-    
+        print('請求失敗，狀態碼:', response.status_code)
+
     # Calculate and print time for each question
     question_end_time = time.time()
     question_duration = question_end_time - question_start_time
-    print(f"QID: {qid} - 花費時間: {question_duration:.2f} 秒")
+    print(f'QID: {qid} - 花費時間: {question_duration:.2f} 秒')
 
 # Calculate and print total time
 total_end_time = time.time()
 total_duration = total_end_time - total_start_time
-print(f"全部題目處理完成，總共花費時間: {total_duration:.2f} 秒")
+print(f'全部題目處理完成，總共花費時間: {total_duration:.2f} 秒')
 
 # Save the output data to a new JSON file
 with open('data/pred_retrieve.json', 'w', encoding='utf-8') as output_file:
     json.dump(output_data, output_file, ensure_ascii=False, indent=4)
 
-print("合併輸出已保存到 pred_retrieve.json 文件中。")
+print('合併輸出已保存到 pred_retrieve.json 文件中。')
diff --git a/src/tools/checkans.py b/src/tools/checkans.py
index 5eeb05a..547700e 100644
--- a/src/tools/checkans.py
+++ b/src/tools/checkans.py
@@ -2,39 +2,39 @@
 from collections import defaultdict
 
 # Load ground truth data
-with open('data/ground_truths_example.json', 'r') as f:
-    ground_truths = json.load(f)["ground_truths"]
+with open('data/ground_truths_example.json') as f:
+    ground_truths = json.load(f)['ground_truths']
 
 # Load predicted data with the new format
-with open('data/pred_retrieve.json', 'r') as f:
-    pred_retrieves = json.load(f)["answers"]
+with open('data/pred_retrieve.json') as f:
+    pred_retrieves = json.load(f)['answers']
 
 # Create a dictionary from predictions for easy lookup
-pred_dict = {item["qid"]: item["retrieve"] for item in pred_retrieves}
+pred_dict = {item['qid']: item['retrieve'] for item in pred_retrieves}
 
 # Initialize counters and data structures
 incorrect_qids = []
 correct_count = 0
-category_counts = defaultdict(lambda: {"correct": 0, "total": 0})
+category_counts = defaultdict(lambda: {'correct': 0, 'total': 0})
 
 # Compare predictions to ground truth
 for ground in ground_truths:
-    qid = ground["qid"]
-    category = ground["category"]
-    correct_retrieve = ground["retrieve"]
+    qid = ground['qid']
+    category = ground['category']
+    correct_retrieve = ground['retrieve']
     predicted_retrieve = pred_dict.get(qid)
 
     if predicted_retrieve == correct_retrieve:
         correct_count += 1
-        category_counts[category]["correct"] += 1
+        category_counts[category]['correct'] += 1
     else:
         incorrect_qids.append(qid)
 
-    category_counts[category]["total"] += 1
+    category_counts[category]['total'] += 1
 
 # Print results
-print("錯誤的題目 QID:", incorrect_qids)
-print(f"總正確題數: {correct_count} / {len(ground_truths)}")
+print('錯誤的題目 QID:', incorrect_qids)
+print(f'總正確題數: {correct_count} / {len(ground_truths)}')
 
 for category, counts in category_counts.items():
     print(f"類別 {category}: {counts['correct']} / {counts['total']}")
diff --git a/src/tools/get_best_alpha.py b/src/tools/get_best_alpha.py
index e6b7527..b2dec99 100644
--- a/src/tools/get_best_alpha.py
+++ b/src/tools/get_best_alpha.py
@@ -1,14 +1,15 @@
-import requests
 import json
 from collections import defaultdict
 
+import requests
+
 # Load questions from the JSON file
-with open('data/questions_example.json', 'r', encoding='utf-8') as file:
+with open('data/questions_example.json', encoding='utf-8') as file:
     questions = json.load(file)['questions']
 
 # Load ground truth data
-with open('data/ground_truths_example.json', 'r', encoding='utf-8') as f:
-    ground_truths = json.load(f)["ground_truths"]
+with open('data/ground_truths_example.json', encoding='utf-8') as f:
+    ground_truths = json.load(f)['ground_truths']
 
 # Dictionary to hold the best alpha and accuracy
 best_alpha = 0.0
@@ -16,60 +17,57 @@
 
 # Loop through alpha values from 0.0 to 1.0
 for alpha in [round(x * 0.1, 1) for x in range(11)]:
-    output_data = {"answers": []}  # Reset output format with "answers" array
+    output_data = {'answers': []}  # Reset output format with "answers" array
 
-    url = "http://127.0.0.1:5000/api/chat"
+    url = 'http://127.0.0.1:5000/api/chat'
 
     # Send each question to the API with the current alpha
     for question in questions:
         # Add the alpha key to the question payload
-        question_with_alpha = {**question, "alpha": alpha}
+        question_with_alpha = {**question, 'alpha': alpha}
 
         # Send POST request
         response = requests.post(url, json=question_with_alpha)
 
         if response.status_code == 200:
             response_json = response.json()
-            qid = question.get("qid")
-            retrieve = response_json.get("retrieve")
+            qid = question.get('qid')
+            retrieve = response_json.get('retrieve')
 
             # Append formatted result to the answers array
-            output_data["answers"].append({
-                "qid": qid,
-                "retrieve": retrieve
-            })
+            output_data['answers'].append({'qid': qid, 'retrieve': retrieve})
         else:
-            print(f"請求失敗，狀態碼: {response.status_code}，Alpha 值: {alpha}")
-    
+            print(f'請求失敗，狀態碼: {response.status_code}，Alpha 值: {alpha}')
+
     # Save predictions for the current alpha
     pred_file = f'data/pred_retrieve_alpha_{alpha}.json'
     with open(pred_file, 'w', encoding='utf-8') as output_file:
         json.dump(output_data, output_file, ensure_ascii=False, indent=4)
 
     # Load predictions for comparison
-    pred_dict = {item["qid"]: item["retrieve"] for item in output_data["answers"]}
+    pred_dict = {item['qid']: item['retrieve'] for item in output_data['answers']}
 
     # Initialize counters and data structures for accuracy calculation
     correct_count = 0
-    category_counts = defaultdict(lambda: {"correct": 0, "total": 0})
+    category_counts = defaultdict(lambda: {'correct': 0, 'total': 0})
 
     # Compare predictions to ground truth
     for ground in ground_truths:
-        qid = ground["qid"]
-        category = ground["category"]
-        correct_retrieve = ground["retrieve"]
+        qid = ground['qid']
+        category = ground['category']
+        correct_retrieve = ground['retrieve']
         predicted_retrieve = pred_dict.get(qid)
 
         if predicted_retrieve == correct_retrieve:
             correct_count += 1
-            category_counts[category]["correct"] += 1
+            category_counts[category]['correct'] += 1
 
-        category_counts[category]["total"] += 1
+        category_counts[category]['total'] += 1
 
     # Calculate accuracy for the current alpha
     accuracy = correct_count / len(ground_truths)
-    print("Corrrect count: ", correct_count)
-    print(f"Alpha: {alpha}, 正確率: {accuracy:.2%}")
+    print('Corrrect count: ', correct_count)
+    print(f'Alpha: {alpha}, 正確率: {accuracy:.2%}')
 
     # Track the best alpha and accuracy
     if accuracy > best_accuracy:
@@ -77,4 +75,4 @@
         best_accuracy = accuracy
 
 # Output the best alpha and accuracy
-print(f"最佳 Alpha 值: {best_alpha}, 準確率: {best_accuracy:.2%}")
+print(f'最佳 Alpha 值: {best_alpha}, 準確率: {best_accuracy:.2%}')
diff --git a/src/utils/ckip.py b/src/utils/ckip.py
index df2f0b8..23d192e 100644
--- a/src/utils/ckip.py
+++ b/src/utils/ckip.py
@@ -1,17 +1,16 @@
-from ckip_transformers.nlp import CkipWordSegmenter, CkipPosTagger
+from ckip_transformers.nlp import CkipPosTagger, CkipWordSegmenter
 
-
-ws_driver = CkipWordSegmenter(model="albert-base")
-pos_driver = CkipPosTagger(model="albert-base")
+ws_driver = CkipWordSegmenter(model='albert-base')
+pos_driver = CkipPosTagger(model='albert-base')
 
 
 def clean(sentence_ws, sentence_pos):
     short_sentence = []
-    stop_pos = set(["Nep", "Nh", "Nb"])
+    stop_pos = set(['Nep', 'Nh', 'Nb'])
     for word_ws, word_pos in zip(sentence_ws, sentence_pos):
-        is_N_or_V = word_pos.startswith("V") or word_pos.startswith("N")
+        is_n_or_v = word_pos.startswith('V') or word_pos.startswith('N')
         is_not_stop_pos = word_pos not in stop_pos
         is_not_one_charactor = not (len(word_ws) == 1)
-        if is_N_or_V and is_not_stop_pos and is_not_one_charactor:
-            short_sentence.append(f"{word_ws}")
-    return " ".join(short_sentence)
+        if is_n_or_v and is_not_stop_pos and is_not_one_charactor:
+            short_sentence.append(f'{word_ws}')
+    return ' '.join(short_sentence)
diff --git a/src/utils/weaviatexreranker.py b/src/utils/weaviatexreranker.py
index f600343..4400670 100644
--- a/src/utils/weaviatexreranker.py
+++ b/src/utils/weaviatexreranker.py
@@ -1,8 +1,10 @@
 import os
+
+import voyageai
 import weaviate
 from langchain.embeddings import OpenAIEmbeddings
+
 import utils.config_log as config_log
-import voyageai
 
 # 載入設定檔案和日誌設定
 config, logger, CONFIG_PATH = config_log.setup_config_and_logging()
@@ -16,6 +18,7 @@
 # 設定 OpenAI API 金鑰
 os.environ['OPENAI_API_KEY'] = config.get('OpenAI', 'api_key')
 
+
 class WeaviateSemanticSearch:
     def __init__(self, classnm):
         self.url = wea_url
@@ -27,9 +30,7 @@ def hybrid_search(self, query, source, num, alpha):
         query_vector = self.embeddings.embed_query(query)
         vector_str = ','.join(map(str, query_vector))
 
-        where_conditions = ' '.join([
-            f'{{path: ["pid"], operator: Equal, valueText: "{pid}"}}' for pid in source
-        ])
+        where_conditions = ' '.join([f'{{path: ["pid"], operator: Equal, valueText: "{pid}"}}' for pid in source])
 
         gql_query = f"""
         {{
@@ -63,7 +64,7 @@ def hybrid_search(self, query, source, num, alpha):
 
 def rerank_with_voyage(query, documents, pids, api_key):
     vo = voyageai.Client(api_key=api_key)
-    reranking = vo.rerank(query, documents, model="rerank-2", top_k=1)
+    reranking = vo.rerank(query, documents, model='rerank-2', top_k=1)
     top_result = reranking.results[0]
 
     # 根據內容找到相對應的 pid
@@ -72,17 +73,17 @@ def rerank_with_voyage(query, documents, pids, api_key):
 
 
 def search_do(question, category, source, alpha):
-    if category == "finance":
-        vdb_named = "Financedev"
-    elif category == "insurance":
-        vdb_named = "Insurancedev"
+    if category == 'finance':
+        vdb_named = 'Financedev'
+    elif category == 'insurance':
+        vdb_named = 'Insurancedev'
     else:
-        vdb_named = "Faqdev"
+        vdb_named = 'Faqdev'
 
     searcher = WeaviateSemanticSearch(vdb_named)
     # 從 Weaviate 取得前 100 筆結果
     top_100_results = searcher.hybrid_search(question, source, 100, alpha=alpha)
-    
+
     # 準備文件和 pid 列表供 rerank 使用
     documents = [result['content'] for result in top_100_results]
     pids = [result['pid'] for result in top_100_results]
@@ -90,8 +91,8 @@ def search_do(question, category, source, alpha):
     # 使用 VoyageAI 重新排序，並取得排名最高的 pid
     top_reranked_result = rerank_with_voyage(question, documents, pids, voyage_api_key)
 
-    print("最相關文件的 PID:")
+    print('最相關文件的 PID:')
     print(f"PID: {top_reranked_result['pid']}")
     print(f"相關性分數: {top_reranked_result['relevance_score']}")
-    
+
     return top_reranked_result['pid']