Add prompts for intent classification

prompts/search/create_search_plan

@@ -1,9 +1,7 @@
 <role>You are a useful search assistant. </role>
 <task>
     You are tasked with creating a comprehensive search plan to explore a complex query given by the user under.
-    Your goal is to break down the query into multiple sub-queries that represents dimensions that must be explored for
-    a holistic understanding of
-    the initial query.
+    Your goal is to break down the query into multiple sub-queries that represents dimensions that must be explored for a holistic understanding of the initial query.
 </task>
 
 <response_format>

prompts/search/determine_intent

@@ -0,0 +1,16 @@
+<role>You are a useful search assistant which can search information through a knowledge graph.</role> 
+<task>
+    You are tasked with determining the user intent behind the user queries. 
+    That is determining what entities are to be investigated and what type of relations should be explored for creating targeted summaries which may answer the user query.
+</task>
+
+<response_format> 
+    For each extracted entity, provide the name as well as the types of relations to consider for answering the user's query.
+    Reply in JSON using the following response schema:
+    {{response_format}}
+</response_format>
+
+<query> 
+    The user query is: 
+    **{{user_query}}** 
+</query>

schemas.py

@@ -33,3 +33,16 @@ class CreateSearchPlanRequest(BaseModel):
 class CreateSearchPlanResponse(BaseModel):
     sub_queries: list[str] = Field(...,
                                    description="A list of subqueries formulated as questions")
+
+
+# ======================================================= Determine intent =========================
+
+class Intent(BaseModel):
+    target_entity: str = Field(
+        ..., description="One of the entities required to fulfil the user intent")
+    target_relations: list[str] = Field(
+        ..., description="A list of one or multiple types of relations required to fulfil the user intent")
+
+
+class ExtractedIntentReponse(BaseModel):
+    intents: list[Intent]