feat(backend): rename keywords to extracted_questions in query pipeline (sub-phase 2.3)

Change QueryDecomposer prompt to generate 2-5 sub-questions instead of keywords. Rename API field from keywords to extracted_questions across models, service, and router.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

backend/app/models/query.py

@@ -10,6 +10,6 @@ class QueryRequest(BaseModel):
 
 
 class QueryResponse(BaseModel):
-    keywords: List[str]
+    extracted_questions: List[str]
     answer: str
     sources: List[SourceMetadata]

backend/app/routers/query.py

@@ -30,13 +30,13 @@ async def query(request: QueryRequest):
 
         logger.info("Query: %s", request.question)
         decomposer = QueryDecomposer(llm_client)
-        keywords = await decomposer.decompose(request.question)
-        logger.info("Keywords: %s", keywords)
+        extracted_questions = await decomposer.decompose(request.question)
+        logger.info("Extracted questions: %s", extracted_questions)
 
-        chunks = rag.retrieve(keywords, n_results=settings.retrieval_n_results)
+        chunks = rag.retrieve(extracted_questions, n_results=settings.retrieval_n_results)
 
         if not chunks:
-            return QueryResponse(keywords=keywords, answer=NO_RESULTS_ANSWER, sources=[])
+            return QueryResponse(extracted_questions=extracted_questions, answer=NO_RESULTS_ANSWER, sources=[])
 
         chunks_for_filter = [(text, meta) for text, meta, _dist in chunks]
         relevance_filter = RelevanceFilter(llm_client)
@@ -45,7 +45,7 @@ async def query(request: QueryRequest):
         )
 
         if not filtered:
-            return QueryResponse(keywords=keywords, answer=NO_RESULTS_ANSWER, sources=[])
+            return QueryResponse(extracted_questions=extracted_questions, answer=NO_RESULTS_ANSWER, sources=[])
 
         chunk_texts = [chunk for chunk, _meta in filtered]
         chunk_metadata = [meta for _chunk, meta in filtered]
@@ -65,7 +65,7 @@ async def query(request: QueryRequest):
             for meta in chunk_metadata
         ]
 
-        return QueryResponse(keywords=keywords, answer=answer, sources=sources)
+        return QueryResponse(extracted_questions=extracted_questions, answer=answer, sources=sources)
 
     except HTTPException:
         raise

backend/app/services/query_decomposer.py

@@ -1,8 +1,8 @@
 """Query decomposer service.
 
 This module provides a lightweight QueryDecomposer that delegates the
-translation of a natural language question into a list of keyword search
-terms to an LLM client.
+decomposition of a natural language question into simplified sub-questions
+to an LLM client.
 """
 from __future__ import annotations
 
@@ -26,11 +26,11 @@ def _extract_json_from_markdown(response: str) -> str:
 
 
 class QueryDecomposer:
-    """Decompose a natural language question into a list of keywords.
+    """Decompose a natural language question into simplified sub-questions.
 
     The class expects an object that exposes an ``async complete(prompt: str) -> str``
     method (an LLM client). The ``decompose`` method builds a prompt, asks the
-    LLM to return a JSON array of strings, and parses that JSON into a Python
+    LLM to return a JSON array of sub-question strings, and parses that JSON into a Python
     list of strings. Edge cases are handled gracefully.
     """
 
@@ -38,20 +38,25 @@ class QueryDecomposer:
         self.llm_client = llm_client
 
     async def decompose(self, question: str) -> List[str]:
-        """Return a list of keywords extracted for the given question.
+        """Return a list of sub-questions extracted for the given question.
 
         Args:
             question: The natural language question to decompose.
 
         Returns:
-            A list of keyword strings. If the LLM response is invalid or the
+            A list of sub-question strings. If the LLM response is invalid or the
             input is empty, an empty list is returned.
         """
 
         if question is None or question.strip() == "":
             return []
 
-        prompt = f"Given question: '{question}', extract key search keywords as JSON array"
+        prompt = (
+            f"Given this question: '{question}'\n\n"
+            f"Break it down into 2-5 simplified sub-questions that would help "
+            f"search for relevant information. Each sub-question should be short "
+            f"and focused on one aspect. Return as a JSON array of strings."
+        )
 
         try:
             response = await self.llm_client.complete(prompt, step_name="QueryDecomposer")