Use LLM-based scoring for platform handler queries (#71)

Platform-retrieved items (Shopify/Wix MCP) were all scoring below
the min_score=51 threshold with the default Pi Labs scorer, resulting
in empty results. Override to 4.1-mini scorer for KeywordRetrievalHandler
queries using the existing override_scoring_provider context manager
(same pattern as aajtak.py).

Pi Labs remains the default scorer for vector DB queries.

Co-authored-by: Prashant Jaiswal <prashjai@Prashants-MacBook-Pro.local>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: 3 people

ask_api/packages/providers/platform_handlers/nlweb_platform_handlers/base.py

@@ -14,6 +14,7 @@
 import logging
 from abc import abstractmethod
 
+from nlweb_core.config import override_scoring_provider
 from nlweb_core.handler import DefaultAskHandler, OutputMethod
 from nlweb_core.protocol.models import AskRequest
 from nlweb_core.retrieved_item import RetrievedItem
@@ -64,16 +65,19 @@ async def _run_query_body(
             num_results=request.query.num_results,
         )
 
-        # Use the same ranking pipeline as DefaultAskHandler
-        final_ranked_answers = await Ranking().rank(
-            items=retrieved_items,
-            query_text=query,
-            item_type=site_config["item_type"],
-            max_results=request.query.max_results,
-            min_score=request.query.min_score,
-            site=request.query.site,
-            start_num=self._get_result_offset(request),
-        )
+        # Use LLM-based scoring for platform-retrieved items.
+        # Platform search APIs return pre-filtered results that need LLM evaluation
+        # rather than the default Pi Labs scorer which under-scores these items.
+        with override_scoring_provider("default", "4.1-mini"):
+            final_ranked_answers = await Ranking().rank(
+                items=retrieved_items,
+                query_text=query,
+                item_type=site_config["item_type"],
+                max_results=request.query.max_results,
+                min_score=request.query.min_score,
+                site=request.query.site,
+                start_num=self._get_result_offset(request),
+            )
 
         await self._send_results(output_method, final_ranked_answers)
         return final_ranked_answers