Fix/document embeddings (#247)

* Update schema for doc embeddings

* Rename embeddings-vectorize to graph-embeddings

* Added document-embeddings processor (broken, needs fixing)

* Added scripts

* Fixed DE queue schema

* Add missing DE process

* Fix doc RAG processing, put graph-rag and doc-rag in appropriate component files.

trustgraph-base/trustgraph/api/api.py

@@ -131,6 +131,35 @@ class Api:
         except:
             raise ProtocolException(f"Response not formatted correctly")
 
+    def document_rag(self, question):
+
+        # The input consists of a question
+        input = {
+            "query": question
+        }
+
+        url = f"{self.url}document-rag"
+
+        # Invoke the API, input is passed as JSON
+        resp = requests.post(url, json=input)
+
+        # Should be a 200 status code
+        if resp.status_code != 200:
+            raise ProtocolException(f"Status code {resp.status_code}")
+
+        try:
+            # Parse the response as JSON
+            object = resp.json()
+        except:
+            raise ProtocolException(f"Expected JSON response")
+
+        self.check_error(resp)
+
+        try:
+            return object["response"]
+        except:
+            raise ProtocolException(f"Response not formatted correctly")
+
     def embeddings(self, text):
 
         # The input consists of a text block

trustgraph-base/trustgraph/clients/document_embeddings_client.py

@@ -38,8 +38,12 @@ class DocumentEmbeddingsClient(BaseClient):
             output_schema=DocumentEmbeddingsResponse,
         )
 
-    def request(self, vectors, limit=10, timeout=300):
+    def request(
+            self, vectors, user="trustgraph", collection="default",
+            limit=10, timeout=300
+    ):
         return self.call(
+            user=user, collection=collection,
             vectors=vectors, limit=limit, timeout=timeout
         ).documents
 

trustgraph-base/trustgraph/schema/documents.py

@@ -35,11 +35,28 @@ chunk_ingest_queue = topic('chunk-load')
 
 ############################################################################
 
+# Document embeddings are embeddings associated with a chunk
+
+class ChunkEmbeddings(Record):
+    chunk = Bytes()
+    vectors = Array(Array(Double()))
+
+# This is a 'batching' mechanism for the above data
+class DocumentEmbeddings(Record):
+    metadata = Metadata()
+    chunks = Array(ChunkEmbeddings())
+
+document_embeddings_store_queue = topic('document-embeddings-store')
+
+############################################################################
+
 # Doc embeddings query
 
 class DocumentEmbeddingsRequest(Record):
     vectors = Array(Array(Double()))
     limit = Integer()
+    user = String()
+    collection = String()
 
 class DocumentEmbeddingsResponse(Record):
     error = Error()