rag: bypass ingestion when payloads are matching already saved data (#…

…157)

Co-authored-by: Avram Tudor <tudor.avram@8x8.com>

skynet/modules/ttt/assistant/v1/models.py

@@ -3,6 +3,8 @@
 
 from pydantic import BaseModel
 
+from skynet.modules.ttt.assistant.constants import assistant_default_system_message
+
 from skynet.modules.ttt.summaries.v1.models import DocumentPayload, HintType
 
 default_max_depth = 5
@@ -25,7 +27,7 @@ class RagPayload(BaseModel):
                 {
                     'urls': ['https://jitsi.github.io/handbook'],
                     'max_depth': default_max_depth,
-                    'system_message': 'You are an AI assistant of Jitsi, a video conferencing platform. You provide response suggestions to the support agent',
+                    'system_message': assistant_default_system_message,
                 }
             ]
         }
@@ -43,7 +45,7 @@ class RagConfig(RagPayload):
                     'error': None,
                     'max_depth': default_max_depth,
                     'status': 'running',
-                    'system_message': 'You are an AI assistant of Jitsi, a video conferencing platform. You provide response suggestions to the support agent',
+                    'system_message': assistant_default_system_message,
                     'urls': ['https://jitsi.github.io/handbook'],
                 }
             ]

skynet/modules/ttt/assistant/v1/router.py

@@ -37,7 +37,7 @@ async def create_rag_db(payload: RagPayload, customer_id=Depends(CustomerId()))
     """
 
     store = await get_vector_store()
-    return await store.create_from_urls(payload, customer_id)
+    return await store.update_from_urls(payload, customer_id)
 
 
 @api_version(1)

skynet/modules/ttt/rag/vector_store.py

@@ -17,6 +17,19 @@
 log = get_logger(__name__)
 
 
+def bypass_ingestion(existing_payload: RagPayload, new_payload: RagPayload):
+    existing_urls = existing_payload.urls
+    new_urls = new_payload.urls
+
+    existing_urls.sort()
+    new_urls.sort()
+
+    return (
+        list(dict.fromkeys(existing_urls)) == list(dict.fromkeys(new_urls))
+        and existing_payload.max_depth == new_payload.max_depth
+    )
+
+
 class SkynetVectorStore(ABC):
     embedding = HuggingFaceEmbeddings(
         model_name=embeddings_model_path, model_kwargs={'device': 'cpu', 'trust_remote_code': True}
@@ -111,13 +124,18 @@ async def workflow(self, payload: RagPayload, store_id: str):
 
         await db.lrem(RUNNING_RAG_KEY, 0, store_id)
 
-    async def create_from_urls(self, payload: RagPayload, store_id: str) -> Optional[RagConfig]:
+    async def update_from_urls(self, payload: RagPayload, store_id: str) -> Optional[RagConfig]:
         """
         Create a vector store with the given id, using the documents crawled from the given URL.
         """
 
+        config = await self.get_config(store_id)
+
         if store_id in await db.lrange(RUNNING_RAG_KEY, 0, -1):
-            return await self.get_config(store_id)
+            return config
+
+        if config and config.status == RagStatus.SUCCESS and bypass_ingestion(config, payload):
+            return await self.update_config(store_id, system_message=payload.system_message)
 
         await db.rpush(RUNNING_RAG_KEY, store_id)
         config = RagConfig(urls=payload.urls, max_depth=payload.max_depth)