обработка вопросов

tg_bot/application/services/rag_service.py

@@ -0,0 +1,139 @@
+import aiohttp
+from tg_bot.infrastructure.external.deepseek_client import DeepSeekClient
+from tg_bot.config.settings import settings
+
+BACKEND_URL = "http://localhost:8001/api/v1"
+
+
+class RAGService:
+
+    def __init__(self):
+        self.deepseek_client = DeepSeekClient()
+
+    async def search_documents_in_collections(
+            self,
+            user_telegram_id: str,
+            query: str,
+            limit_per_collection: int = 5
+    ) -> list[dict]:
+        try:
+            async with aiohttp.ClientSession() as session:
+                async with session.get(
+                        f"{BACKEND_URL}/users/telegram/{user_telegram_id}"
+                ) as user_response:
+                    if user_response.status != 200:
+                        return []
+
+                    user_data = await user_response.json()
+                    user_uuid = str(user_data.get("user_id"))
+
+                    if not user_uuid:
+                        return []
+
+                async with session.get(
+                        f"{BACKEND_URL}/collections/",
+                        headers={"X-Telegram-ID": user_telegram_id}
+                ) as collections_response:
+                    if collections_response.status != 200:
+                        return []
+
+                    collections = await collections_response.json()
+
+            all_documents = []
+            for collection in collections:
+                collection_id = collection.get("collection_id")
+                if not collection_id:
+                    continue
+
+                try:
+                    async with aiohttp.ClientSession() as search_session:
+                        async with search_session.get(
+                                f"{BACKEND_URL}/documents/collection/{collection_id}",
+                                params={"search": query, "limit": limit_per_collection},
+                                headers={"X-Telegram-ID": user_telegram_id}
+                        ) as search_response:
+                            if search_response.status == 200:
+                                documents = await search_response.json()
+                                for doc in documents:
+                                    doc["collection_name"] = collection.get("name", "Unknown")
+                                    all_documents.append(doc)
+                except Exception as e:
+                    print(f"Error searching collection {collection_id}: {e}")
+                    continue
+
+            return all_documents[:20]
+
+        except Exception as e:
+            print(f"Error searching documents: {e}")
+            return []
+
+    async def generate_answer_with_rag(
+            self,
+            question: str,
+            user_telegram_id: str
+    ) -> dict:
+        documents = await self.search_documents_in_collections(
+            user_telegram_id,
+            question
+        )
+
+        context_parts = []
+        sources = []
+
+        for doc in documents[:5]:
+            title = doc.get("title", "Без названия")
+            content = doc.get("content", "")[:1000]
+            collection_name = doc.get("collection_name", "Unknown")
+
+            context_parts.append(f"Документ: {title}\nКоллекция: {collection_name}\nСодержание: {content[:500]}...")
+            sources.append({
+                "title": title,
+                "collection": collection_name,
+                "document_id": doc.get("document_id")
+            })
+
+        context = "\n\n".join(context_parts) if context_parts else "Релевантные документы не найдены."
+
+        system_prompt = """Ты - помощник-юрист, который отвечает на вопросы на основе предоставленных документов.
+Используй информацию из документов для формирования точного и полезного ответа.
+Если в документах нет информации для ответа, честно скажи об этом."""
+
+        user_prompt = f"""Контекст из документов:
+{context}
+
+Вопрос пользователя: {question}
+
+Ответь на вопрос, используя информацию из предоставленных документов. Если информации недостаточно, укажи это."""
+
+        try:
+            messages = [
+                {"role": "system", "content": system_prompt},
+                {"role": "user", "content": user_prompt}
+            ]
+
+            response = await self.deepseek_client.chat_completion(
+                messages=messages,
+                temperature=0.7,
+                max_tokens=2000
+            )
+
+            return {
+                "answer": response.get("content", "Failed to generate answer"),
+                "sources": sources,
+                "usage": response.get("usage", {})
+            }
+
+        except Exception as e:
+            print(f"Error generating answer: {e}")
+            if documents:
+                return {
+                    "answer": f"Found {len(documents)} documents but failed to generate answer",
+                    "sources": sources[:3],
+                    "usage": {}
+                }
+            else:
+                return {
+                    "answer": "No relevant documents found",
+                    "sources": [],
+                    "usage": {}
+                }