Update utils.py

utils.py

@@ -1,274 +1,290 @@
-import logging
-import sys
-from llama_index.llms.google_genai import GoogleGenAI
-from llama_index.llms.openai import OpenAI
-from llama_index.embeddings.huggingface import HuggingFaceEmbedding
-from sentence_transformers import CrossEncoder
-from config import AVAILABLE_MODELS, DEFAULT_MODEL, GOOGLE_API_KEY
-import time
-from index_retriever import rerank_nodes
-from my_logging import log_message
-from config import PROMPT_SIMPLE_POISK
-
-def get_llm_model(model_name):
-    try:
-        model_config = AVAILABLE_MODELS.get(model_name)
-        if not model_config:
-            log_message(f"Модель {model_name} не найдена, использую модель по умолчанию")
-            model_config = AVAILABLE_MODELS[DEFAULT_MODEL]
-        
-        if not model_config.get("api_key"):
-            raise Exception(f"API ключ не найден для модели {model_name}")
-        
-        if model_config["provider"] == "google":
-            return GoogleGenAI(
-                model=model_config["model_name"], 
-                api_key=model_config["api_key"]
-            )
-        elif model_config["provider"] == "openai":
-            return OpenAI(
-                model=model_config["model_name"],
-                api_key=model_config["api_key"]
-            )
-        else:
-            raise Exception(f"Неподдерживаемый провайдер: {model_config['provider']}")
-            
-    except Exception as e:
-        log_message(f"Ошибка создания модели {model_name}: {str(e)}")
-        return GoogleGenAI(model="gemini-2.0-flash", api_key=GOOGLE_API_KEY)
-
-def get_embedding_model(model_name="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"):
-    return HuggingFaceEmbedding(model_name=model_name)
-
-def get_reranker_model(model_name='cross-encoder/ms-marco-MiniLM-L-12-v2'):
-    return CrossEncoder(model_name)
-
-def generate_sources_html(nodes, chunks_df=None):
-    html = "<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; max-height: 400px; overflow-y: auto;'>"
-    html += "<h3 style='color: #63b3ed; margin-top: 0;'>Источники:</h3>"
-    
-    sources_by_doc = {}
-    
-    for i, node in enumerate(nodes):
-        metadata = node.metadata if hasattr(node, 'metadata') else {}
-        doc_type = metadata.get('type', 'text')
-        doc_id = metadata.get('document_id', 'unknown')
-        
-        if doc_type == 'table' or doc_type == 'table_row':
-            table_num = metadata.get('table_number', 'unknown')
-            key = f"{doc_id}_table_{table_num}"
-        elif doc_type == 'image':
-            image_num = metadata.get('image_number', 'unknown')
-            key = f"{doc_id}_image_{image_num}"
-        else:
-            section_path = metadata.get('section_path', '')
-            section_id = metadata.get('section_id', '')
-            section_key = section_path if section_path else section_id
-            key = f"{doc_id}_text_{section_key}"
-        
-        if key not in sources_by_doc:
-            sources_by_doc[key] = {
-                'doc_id': doc_id,
-                'doc_type': doc_type,
-                'metadata': metadata,
-                'sections': set()
-            }
-        
-        if doc_type not in ['table', 'table_row', 'image']:
-            section_path = metadata.get('section_path', '')
-            section_id = metadata.get('section_id', '')
-            if section_path:
-                sources_by_doc[key]['sections'].add(f"пункт {section_path}")
-            elif section_id and section_id != 'unknown':
-                sources_by_doc[key]['sections'].add(f"пункт {section_id}")
-    
-    for source_info in sources_by_doc.values():
-        metadata = source_info['metadata']
-        doc_type = source_info['doc_type']
-        doc_id = source_info['doc_id']
-        
-        html += f"<div style='margin-bottom: 15px; padding: 15px; border: 1px solid #4a5568; border-radius: 8px; background-color: #1a202c;'>"
-        
-        if doc_type == 'text':
-            html += f"<h4 style='margin: 0 0 10px 0; color: #63b3ed;'>📄 {doc_id}</h4>"
-        elif doc_type == 'table' or doc_type == 'table_row':
-            table_num = metadata.get('table_number', 'unknown')
-            table_title = metadata.get('table_title', '')
-            if table_num and table_num != 'unknown':
-                if not str(table_num).startswith('№'):
-                    table_num = f"№{table_num}"
-                html += f"<h4 style='margin: 0 0 10px 0; color: #68d391;'>📊 Таблица {table_num} - {doc_id}</h4>"
-                if table_title and table_title != 'unknown':
-                    html += f"<p style='margin: 5px 0; color: #a0aec0; font-size: 14px;'>{table_title}</p>"
-            else:
-                html += f"<h4 style='margin: 0 0 10px 0; color: #68d391;'>📊 Таблица - {doc_id}</h4>"
-        elif doc_type == 'image':
-            image_num = metadata.get('image_number', 'unknown')
-            image_title = metadata.get('image_title', '')
-            if image_num and image_num != 'unknown':
-                if not str(image_num).startswith('№'):
-                    image_num = f"№{image_num}"
-                html += f"<h4 style='margin: 0 0 10px 0; color: #fbb6ce;'>🖼️ Изображение {image_num} - {doc_id}</h4>"
-                if image_title and image_title != 'unknown':
-                    html += f"<p style='margin: 5px 0; color: #a0aec0; font-size: 14px;'>{image_title}</p>"
-        
-        if chunks_df is not None and 'file_link' in chunks_df.columns and doc_type == 'text':
-            doc_rows = chunks_df[chunks_df['document_id'] == doc_id]
-            if not doc_rows.empty:
-                file_link = doc_rows.iloc[0]['file_link']
-                html += f"<a href='{file_link}' target='_blank' style='color: #68d391; text-decoration: none; font-size: 14px; display: inline-block; margin-top: 10px;'>🔗 Ссылка на документ</a><br>"
-        
-        html += "</div>"
-    
-    html += "</div>"
-    return html
-
-def deduplicate_nodes(nodes):
-    """Deduplicate retrieved nodes based on content and metadata"""
-    seen = set()
-    unique_nodes = []
-    
-    for node in nodes:
-        doc_id = node.metadata.get('document_id', '')
-        node_type = node.metadata.get('type', 'text')
-        
-        if node_type == 'table' or node_type == 'table_row':
-            table_num = node.metadata.get('table_number', '')
-            table_identifier = node.metadata.get('table_identifier', table_num)
-            
-            # Use row range to distinguish table chunks
-            row_start = node.metadata.get('row_start', '')
-            row_end = node.metadata.get('row_end', '')
-            is_complete = node.metadata.get('is_complete_table', False)
-            
-            if is_complete:
-                identifier = f"{doc_id}|table|{table_identifier}|complete"
-            elif row_start != '' and row_end != '':
-                identifier = f"{doc_id}|table|{table_identifier}|rows_{row_start}_{row_end}"
-            else:
-                # Fallback: use chunk_id if available
-                chunk_id = node.metadata.get('chunk_id', '')
-                if chunk_id != '':
-                    identifier = f"{doc_id}|table|{table_identifier}|chunk_{chunk_id}"
-                else:
-                    # Last resort: hash first 100 chars of content
-                    import hashlib
-                    content_hash = hashlib.md5(node.text[:100].encode()).hexdigest()[:8]
-                    identifier = f"{doc_id}|table|{table_identifier}|{content_hash}"
-                    
-        elif node_type == 'image':
-            img_num = node.metadata.get('image_number', '')
-            identifier = f"{doc_id}|image|{img_num}"
-            
-        else:  # text
-            section_id = node.metadata.get('section_id', '')
-            chunk_id = node.metadata.get('chunk_id', 0)
-            # For text, section_id + chunk_id should be unique
-            identifier = f"{doc_id}|text|{section_id}|{chunk_id}"
-        
-        if identifier not in seen:
-            seen.add(identifier)
-            unique_nodes.append(node)
-    
-    return unique_nodes
-
-def debug_search_tables(vector_index, search_term="С-25"):
-    """Debug function to find all tables containing a specific term"""
-    all_nodes = list(vector_index.docstore.docs.values())
-    
-    matching = []
-    for node in all_nodes:
-        if node.metadata.get('type') == 'table':
-            text = node.get_content()
-            if search_term in text or search_term in node.metadata.get('table_title', ''):
-                matching.append({
-                    'doc_id': node.metadata.get('document_id'),
-                    'table_num': node.metadata.get('table_number'),
-                    'title': node.metadata.get('table_title', '')[:100]
-                })
-    
-    log_message(f"\n{'='*60}")
-    log_message(f"DEBUG: Found {len(matching)} tables containing '{search_term}'")
-    for m in matching:
-        log_message(f"  • {m['doc_id']} - Table {m['table_num']}: {m['title']}")
-    log_message(f"{'='*60}\n")
-    
-    return matching
-
-# Add this import at the top of utils.py
-from documents_prep import normalize_text
-
-# MODIFIED: Update answer_question function
-def answer_question(question, query_engine, reranker, current_model, chunks_df=None):
-    # NORMALIZE the question to convert C to С
-    normalized_question = normalize_text(question)
-    
-    if query_engine is None:
-        return "<div style='background-color: #e53e3e; color: white; padding: 20px; border-radius: 10px;'>Система не инициализирована</div>", "", ""
-    
-    try:
-        start_time = time.time()
-        # Use NORMALIZED question for retrieval
-        retrieved_nodes = query_engine.retriever.retrieve(normalized_question)
-        log_message(f"user query: {question}")
-        log_message(f"normalized query: {normalized_question}")
-
-        
-        log_message(f"RETRIEVED: {len(retrieved_nodes)} nodes")
-        
-        unique_retrieved = deduplicate_nodes(retrieved_nodes)
-
-        # DEBUG: Log what was retrieved
-        log_message(f"RETRIEVED: unique {len(unique_retrieved)} nodes")
-        for i, node in enumerate(unique_retrieved):  # All debug
-            table_num = node.metadata.get('table_number', 'N/A')
-            table_title = node.metadata.get('table_title', 'N/A')
-            doc_id = node.metadata.get('document_id', 'N/A')
-            log_message(f"  [{i+1}] {doc_id} - Table {table_num}: {table_title[:50]}")
-        log_message(f"UNIQUE NODES: {len(unique_retrieved)} nodes")
-        
-        # Simple reranking with NORMALIZED question
-        reranked_nodes = rerank_nodes(normalized_question, unique_retrieved, reranker, top_k=20)
-        
-        # Direct query without formatting - use normalized question
-        response = query_engine.query(normalized_question)
-        
-        end_time = time.time()
-        processing_time = end_time - start_time
-        
-        log_message(f"Обработка завершена за {processing_time:.2f}с")
-        
-        sources_html = generate_sources_html(reranked_nodes, chunks_df)
-        
-        answer_with_time = f"""<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; margin-bottom: 10px;'>
-        <h3 style='color: #63b3ed; margin-top: 0;'>Ответ (Модель: {current_model}):</h3>
-        <div style='line-height: 1.6; font-size: 16px;'>{response.response}</div>
-        <div style='margin-top: 15px; padding-top: 10px; border-top: 1px solid #4a5568; font-size: 14px; color: #a0aec0;'>
-        Время обработки: {processing_time:.2f} секунд
-        </div>
-        </div>"""
-        log_message(f"Model Answer: {response.response}")
-        
-        chunk_info = []
-        for node in reranked_nodes:
-            metadata = node.metadata if hasattr(node, 'metadata') else {}
-            chunk_info.append({
-                'document_id': metadata.get('document_id', 'unknown'),
-                'section_id': metadata.get('section_id', 'unknown'),
-                'section_path': metadata.get('section_path', ''),
-                'section_text': metadata.get('section_text', ''),
-                'type': metadata.get('type', 'text'),
-                'table_number': metadata.get('table_number', ''),
-                'image_number': metadata.get('image_number', ''),
-                'chunk_size': len(node.text),
-                'chunk_text': node.text
-            })
-        from app import create_chunks_display_html
-        chunks_html = create_chunks_display_html(chunk_info)
-
-        return answer_with_time, sources_html, chunks_html
-        
-    except Exception as e:
-        log_message(f"Ошибка: {str(e)}")
-        error_msg = f"<div style='background-color: #e53e3e; color: white; padding: 20px; border-radius: 10px;'>Ошибка: {str(e)}</div>"
+import logging
+import sys
+from llama_index.llms.google_genai import GoogleGenAI
+from llama_index.llms.openai import OpenAI
+from llama_index.embeddings.huggingface import HuggingFaceEmbedding
+from sentence_transformers import CrossEncoder
+from config import AVAILABLE_MODELS, DEFAULT_MODEL, GOOGLE_API_KEY
+import time
+from index_retriever import rerank_nodes
+from my_logging import log_message
+from config import PROMPT_SIMPLE_POISK
+
+def get_llm_model(model_name):
+    try:
+        model_config = AVAILABLE_MODELS.get(model_name)
+        if not model_config:
+            log_message(f"Модель {model_name} не найдена, использую модель по умолчанию")
+            model_config = AVAILABLE_MODELS[DEFAULT_MODEL]
+        
+        if not model_config.get("api_key"):
+            raise Exception(f"API ключ не найден для модели {model_name}")
+        
+        if model_config["provider"] == "google":
+            return GoogleGenAI(
+                model=model_config["model_name"], 
+                api_key=model_config["api_key"]
+            )
+        elif model_config["provider"] == "openai":
+            return OpenAI(
+                model=model_config["model_name"],
+                api_key=model_config["api_key"]
+            )
+        else:
+            raise Exception(f"Неподдерживаемый провайдер: {model_config['provider']}")
+            
+    except Exception as e:
+        log_message(f"Ошибка создания модели {model_name}: {str(e)}")
+        return GoogleGenAI(model="gemini-2.0-flash", api_key=GOOGLE_API_KEY)
+
+def get_embedding_model(model_name="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"):
+    return HuggingFaceEmbedding(model_name=model_name)
+
+def get_reranker_model(model_name='cross-encoder/ms-marco-MiniLM-L-12-v2'):
+    return CrossEncoder(model_name)
+
+def generate_sources_html(nodes, chunks_df=None):
+    html = "<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; max-height: 400px; overflow-y: auto;'>"
+    html += "<h3 style='color: #63b3ed; margin-top: 0;'>Источники:</h3>"
+    
+    sources_by_doc = {}
+    
+    for i, node in enumerate(nodes):
+        metadata = node.metadata if hasattr(node, 'metadata') else {}
+        doc_type = metadata.get('type', 'text')
+        doc_id = metadata.get('document_id', 'unknown')
+        
+        if doc_type == 'table' or doc_type == 'table_row':
+            table_num = metadata.get('table_number', 'unknown')
+            key = f"{doc_id}_table_{table_num}"
+        elif doc_type == 'image':
+            image_num = metadata.get('image_number', 'unknown')
+            key = f"{doc_id}_image_{image_num}"
+        else:
+            section_path = metadata.get('section_path', '')
+            section_id = metadata.get('section_id', '')
+            section_key = section_path if section_path else section_id
+            key = f"{doc_id}_text_{section_key}"
+        
+        if key not in sources_by_doc:
+            sources_by_doc[key] = {
+                'doc_id': doc_id,
+                'doc_type': doc_type,
+                'metadata': metadata,
+                'sections': set()
+            }
+        
+        if doc_type not in ['table', 'table_row', 'image']:
+            section_path = metadata.get('section_path', '')
+            section_id = metadata.get('section_id', '')
+            if section_path:
+                sources_by_doc[key]['sections'].add(f"пункт {section_path}")
+            elif section_id and section_id != 'unknown':
+                sources_by_doc[key]['sections'].add(f"пункт {section_id}")
+    
+    for source_info in sources_by_doc.values():
+        metadata = source_info['metadata']
+        doc_type = source_info['doc_type']
+        doc_id = source_info['doc_id']
+        
+        html += f"<div style='margin-bottom: 15px; padding: 15px; border: 1px solid #4a5568; border-radius: 8px; background-color: #1a202c;'>"
+        
+        if doc_type == 'text':
+            html += f"<h4 style='margin: 0 0 10px 0; color: #63b3ed;'>📄 {doc_id}</h4>"
+        elif doc_type == 'table' or doc_type == 'table_row':
+            table_num = metadata.get('table_number', 'unknown')
+            table_title = metadata.get('table_title', '')
+            if table_num and table_num != 'unknown':
+                if not str(table_num).startswith('№'):
+                    table_num = f"№{table_num}"
+                html += f"<h4 style='margin: 0 0 10px 0; color: #68d391;'>📊 Таблица {table_num} - {doc_id}</h4>"
+                if table_title and table_title != 'unknown':
+                    html += f"<p style='margin: 5px 0; color: #a0aec0; font-size: 14px;'>{table_title}</p>"
+            else:
+                html += f"<h4 style='margin: 0 0 10px 0; color: #68d391;'>📊 Таблица - {doc_id}</h4>"
+        elif doc_type == 'image':
+            image_num = metadata.get('image_number', 'unknown')
+            image_title = metadata.get('image_title', '')
+            if image_num and image_num != 'unknown':
+                if not str(image_num).startswith('№'):
+                    image_num = f"№{image_num}"
+                html += f"<h4 style='margin: 0 0 10px 0; color: #fbb6ce;'>🖼️ Изображение {image_num} - {doc_id}</h4>"
+                if image_title and image_title != 'unknown':
+                    html += f"<p style='margin: 5px 0; color: #a0aec0; font-size: 14px;'>{image_title}</p>"
+        
+        if chunks_df is not None and 'file_link' in chunks_df.columns and doc_type == 'text':
+            doc_rows = chunks_df[chunks_df['document_id'] == doc_id]
+            if not doc_rows.empty:
+                file_link = doc_rows.iloc[0]['file_link']
+                html += f"<a href='{file_link}' target='_blank' style='color: #68d391; text-decoration: none; font-size: 14px; display: inline-block; margin-top: 10px;'>🔗 Ссылка на документ</a><br>"
+        
+        html += "</div>"
+    
+    html += "</div>"
+    return html
+
+def deduplicate_nodes(nodes):
+    """Deduplicate retrieved nodes based on content and metadata"""
+    seen = set()
+    unique_nodes = []
+    
+    for node in nodes:
+        doc_id = node.metadata.get('document_id', '')
+        node_type = node.metadata.get('type', 'text')
+        
+        if node_type == 'table' or node_type == 'table_row':
+            table_num = node.metadata.get('table_number', '')
+            table_identifier = node.metadata.get('table_identifier', table_num)
+            
+            # Use row range to distinguish table chunks
+            row_start = node.metadata.get('row_start', '')
+            row_end = node.metadata.get('row_end', '')
+            is_complete = node.metadata.get('is_complete_table', False)
+            
+            if is_complete:
+                identifier = f"{doc_id}|table|{table_identifier}|complete"
+            elif row_start != '' and row_end != '':
+                identifier = f"{doc_id}|table|{table_identifier}|rows_{row_start}_{row_end}"
+            else:
+                # Fallback: use chunk_id if available
+                chunk_id = node.metadata.get('chunk_id', '')
+                if chunk_id != '':
+                    identifier = f"{doc_id}|table|{table_identifier}|chunk_{chunk_id}"
+                else:
+                    # Last resort: hash first 100 chars of content
+                    import hashlib
+                    content_hash = hashlib.md5(node.text[:100].encode()).hexdigest()[:8]
+                    identifier = f"{doc_id}|table|{table_identifier}|{content_hash}"
+                    
+        elif node_type == 'image':
+            img_num = node.metadata.get('image_number', '')
+            identifier = f"{doc_id}|image|{img_num}"
+            
+        else:  # text
+            section_id = node.metadata.get('section_id', '')
+            chunk_id = node.metadata.get('chunk_id', 0)
+            # For text, section_id + chunk_id should be unique
+            identifier = f"{doc_id}|text|{section_id}|{chunk_id}"
+        
+        if identifier not in seen:
+            seen.add(identifier)
+            unique_nodes.append(node)
+    
+    return unique_nodes
+
+def debug_search_tables(vector_index, search_term="С-25"):
+    """Debug function to find all tables containing a specific term"""
+    all_nodes = list(vector_index.docstore.docs.values())
+    
+    matching = []
+    for node in all_nodes:
+        if node.metadata.get('type') == 'table':
+            text = node.get_content()
+            if search_term in text or search_term in node.metadata.get('table_title', ''):
+                matching.append({
+                    'doc_id': node.metadata.get('document_id'),
+                    'table_num': node.metadata.get('table_number'),
+                    'title': node.metadata.get('table_title', '')[:100]
+                })
+    
+    log_message(f"\n{'='*60}")
+    log_message(f"DEBUG: Found {len(matching)} tables containing '{search_term}'")
+    for m in matching:
+        log_message(f"  • {m['doc_id']} - Table {m['table_num']}: {m['title']}")
+    log_message(f"{'='*60}\n")
+    
+    return matching
+
+from documents_prep import normalize_text, normalize_steel_designations
+
+def answer_question(question, query_engine, reranker, current_model, chunks_df=None, rerank_top_k=20):
+    
+    normalized_question = normalize_text(question)
+    log_message(f"Normalized question: {normalized_question}")
+    normalized_question_2, query_changes, change_list = normalize_steel_designations(question)  # FIX: 3 values
+    log_message(f"After steel normalization: {normalized_question_2}")
+    if change_list:
+        log_message(f"Query changes: {', '.join(change_list)}")    
+    if query_engine is None:
+        return "<div style='background-color: #e53e3e; color: white; padding: 20px; border-radius: 10px;'>Система не инициализирована</div>", "", ""
+    
+    try:
+        start_time = time.time()
+        retrieved_nodes = query_engine.retriever.retrieve(normalized_question_2)
+        log_message(f"user query: {question}")
+        log_message(f"normalized query: {normalized_question}")
+        log_message(f"after steel normalization: {normalized_question_2}")
+        log_message(f"Steel grades normalized in query: {query_changes}")
+
+        
+        log_message(f"RETRIEVED: {len(retrieved_nodes)} nodes")
+        
+        unique_retrieved = deduplicate_nodes(retrieved_nodes)
+
+        # IMPROVED DEBUG: Log what was actually retrieved with FULL metadata
+        log_message(f"RETRIEVED: unique {len(unique_retrieved)} nodes")
+        for i, node in enumerate(unique_retrieved):
+            node_type = node.metadata.get('type', 'text')
+            doc_id = node.metadata.get('document_id', 'N/A')
+            
+            if node_type == 'table':
+                table_num = node.metadata.get('table_number', 'N/A')
+                table_id = node.metadata.get('table_identifier', 'N/A')
+                table_title = node.metadata.get('table_title', 'N/A')
+                # Show first 200 chars of content to verify it's the right table
+                content_preview = node.text[:200].replace('\n', ' ')
+                log_message(f"  [{i+1}] {doc_id} - Table {table_num} | ID: {table_id}")
+                log_message(f"      Title: {table_title[:80]}")
+                log_message(f"      Content: {content_preview}...")
+            else:
+                section = node.metadata.get('section_id', 'N/A')
+                log_message(f"  [{i+1}] {doc_id} - Text section {section}")
+        
+        log_message(f"UNIQUE NODES: {len(unique_retrieved)} nodes")
+        
+        # Simple reranking with NORMALIZED question and PARAMETERIZED top_k
+        reranked_nodes = rerank_nodes(normalized_question, unique_retrieved, reranker, 
+                                     top_k=rerank_top_k)  # NOW PARAMETERIZED
+        
+        # Direct query without formatting - use normalized question
+        response = query_engine.query(normalized_question)
+        
+        end_time = time.time()
+        processing_time = end_time - start_time
+        
+        log_message(f"Обработка завершена за {processing_time:.2f}с")
+        
+        sources_html = generate_sources_html(reranked_nodes, chunks_df)
+        
+        answer_with_time = f"""<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; margin-bottom: 10px;'>
+        <h3 style='color: #63b3ed; margin-top: 0;'>Ответ (Модель: {current_model}):</h3>
+        <div style='line-height: 1.6; font-size: 16px;'>{response.response}</div>
+        <div style='margin-top: 15px; padding-top: 10px; border-top: 1px solid #4a5568; font-size: 14px; color: #a0aec0;'>
+        Время обработки: {processing_time:.2f} секунд 
+        </div>
+        </div>"""
+        log_message(f"Model Answer: {response.response}")
+        
+        chunk_info = []
+        for node in reranked_nodes:
+            metadata = node.metadata if hasattr(node, 'metadata') else {}
+            chunk_info.append({
+                'document_id': metadata.get('document_id', 'unknown'),
+                'section_id': metadata.get('section_id', 'unknown'),
+                'section_path': metadata.get('section_path', ''),
+                'section_text': metadata.get('section_text', ''),
+                'type': metadata.get('type', 'text'),
+                'table_number': metadata.get('table_number', ''),
+                'image_number': metadata.get('image_number', ''),
+                'chunk_size': len(node.text),
+                'chunk_text': node.text
+            })
+        from app import create_chunks_display_html
+        chunks_html = create_chunks_display_html(chunk_info)
+
+        return answer_with_time, sources_html, chunks_html
+        
+    except Exception as e:
+        log_message(f"Ошибка: {str(e)}")
+        error_msg = f"<div style='background-color: #e53e3e; color: white; padding: 20px; border-radius: 10px;'>Ошибка: {str(e)}</div>"
         return error_msg, "", ""