route group message extraction through AIOrchestrator for HuggingFace fallback on Groq rate limits

app/ai/orchestrator.py

@@ -27,6 +27,42 @@ class AIOrchestrator:
         self.temperature = temperature
         self.max_tool_iterations = max_tool_iterations
 
+    async def chat(
+        self,
+        *,
+        messages: list[dict[str, Any]],
+        tools: list[dict[str, Any]] | None = None,
+        tool_choice: str | dict[str, Any] | None = None,
+        temperature: float = 0.2,
+    ) -> AIProviderResponse:
+        try:
+            return await self.primary.chat(
+                messages,
+                tools=tools,
+                tool_choice=tool_choice,
+                temperature=temperature,
+            )
+        except InvalidToolCallGenerationError:
+            logger.warning("Primary provider generated invalid tool call; retrying once")
+            try:
+                return await self.primary.chat(
+                    messages,
+                    tools=tools,
+                    tool_choice=tool_choice,
+                    temperature=max(temperature - 0.2, 0.1),
+                )
+            except RetryableProviderError:
+                logger.warning("Primary retry failed; falling back to Hugging Face")
+        except RetryableProviderError:
+            logger.warning("Primary provider failed; falling back to Hugging Face")
+
+        return await self.fallback.chat(
+            messages,
+            tools=tools,
+            tool_choice=tool_choice,
+            temperature=temperature,
+        )
+
     async def generate_reply(
         self,
         *,

app/services/container.py

@@ -44,7 +44,7 @@ class ServiceContainer:
         group_message = GroupMessageService(
             repository=repository,
             embeddings=embeddings,
-            provider=ai.primary,
+            ai=ai,
             settings=settings,
         )
         admin = AdminService(repository=repository, embeddings=embeddings, settings=settings)

app/services/group_message_service.py

@@ -4,7 +4,7 @@ from datetime import date
 from pathlib import Path
 from typing import Any
 
-from app.ai.providers import ChatProvider
+from app.ai.orchestrator import AIOrchestrator
 from app.config import Settings
 from app.database.supabase import SupabaseRepository
 from app.models.domain import ExtractedTrip, WhatsAppInboundMessage
@@ -24,12 +24,12 @@ class GroupMessageService:
         *,
         repository: SupabaseRepository,
         embeddings: JinaEmbeddingService,
-        provider: ChatProvider,
+        ai: AIOrchestrator,
         settings: Settings,
     ) -> None:
         self.repository = repository
         self.embeddings = embeddings
-        self.provider = provider
+        self.ai = ai
         self.settings = settings
 
     async def handle_group_message(
@@ -153,7 +153,7 @@ class GroupMessageService:
         prompt = prompt_template.format(current_datetime=dt.isoformat())
 
         try:
-            response = await self.provider.chat(
+            response = await self.ai.chat(
                 messages=[
                     {"role": "system", "content": prompt},
                     {"role": "user", "content": text},

tests/test_ai_orchestrator.py

@@ -137,3 +137,70 @@ async def test_ai_reports_invalid_tool_arguments_to_model():
 
     assert reply == "Please share the question again."
     assert "Invalid tool arguments" in primary.calls[1]["messages"][-1]["content"]
+
+
+@pytest.mark.asyncio
+async def test_chat_falls_back_when_primary_rate_limited():
+    primary = ScriptedProvider("groq", [RetryableProviderError("rate limited")])
+    fallback = ScriptedProvider("huggingface", [AIProviderResponse(content="fallback reply")])
+    orchestrator = AIOrchestrator(
+        primary=primary,
+        fallback=fallback,
+        temperature=0.2,
+        max_tool_iterations=3,
+    )
+
+    response = await orchestrator.chat(
+        messages=[{"role": "user", "content": "test"}],
+    )
+
+    assert response.content == "fallback reply"
+    assert len(primary.calls) == 1
+    assert len(fallback.calls) == 1
+
+
+@pytest.mark.asyncio
+async def test_chat_returns_primary_on_success():
+    primary = ScriptedProvider("groq", [AIProviderResponse(content="primary reply")])
+    fallback = ScriptedProvider("huggingface", [AIProviderResponse(content="fallback")])
+    orchestrator = AIOrchestrator(
+        primary=primary,
+        fallback=fallback,
+        temperature=0.2,
+        max_tool_iterations=3,
+    )
+
+    response = await orchestrator.chat(
+        messages=[{"role": "user", "content": "test"}],
+    )
+
+    assert response.content == "primary reply"
+    assert len(primary.calls) == 1
+    assert len(fallback.calls) == 0
+
+
+@pytest.mark.asyncio
+async def test_chat_retries_invalid_tool_call_before_fallback():
+    primary = ScriptedProvider(
+        "groq",
+        [
+            InvalidToolCallGenerationError("bad tool"),
+            AIProviderResponse(content="primary retry reply"),
+        ],
+    )
+    fallback = ScriptedProvider("huggingface", [AIProviderResponse(content="fallback")])
+    orchestrator = AIOrchestrator(
+        primary=primary,
+        fallback=fallback,
+        temperature=0.4,
+        max_tool_iterations=3,
+    )
+
+    response = await orchestrator.chat(
+        messages=[{"role": "user", "content": "test"}],
+        temperature=0.4,
+    )
+
+    assert response.content == "primary retry reply"
+    assert [call["temperature"] for call in primary.calls] == [0.4, 0.2]
+    assert fallback.calls == []

tests/test_group_message_service.py

@@ -96,7 +96,7 @@ async def test_non_trip_message_is_ignored(settings: Settings) -> None:
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -117,7 +117,7 @@ async def test_trip_ad_from_new_driver_creates_all_entities(settings: Settings)
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -153,7 +153,7 @@ async def test_trip_ad_from_existing_customer_is_discarded(settings: Settings) -
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -180,7 +180,7 @@ async def test_existing_unregistered_driver_adds_new_trip(settings: Settings) ->
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -221,7 +221,7 @@ async def test_existing_unregistered_driver_duplicate_trip_is_skipped(settings:
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -250,7 +250,7 @@ async def test_incomplete_fields_cause_skip(settings: Settings) -> None:
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -278,7 +278,7 @@ async def test_missing_phone_cause_skip(settings: Settings) -> None:
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -297,7 +297,7 @@ async def test_duplicate_group_message_is_deduplicated(settings: Settings) -> No
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -320,7 +320,7 @@ async def test_phone_normalization(settings: Settings) -> None:
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -341,7 +341,7 @@ async def test_invalid_json_response_cause_skip(settings: Settings) -> None:
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -361,7 +361,7 @@ async def test_markdown_fenced_json_is_parsed(settings: Settings) -> None:
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -382,7 +382,7 @@ async def test_trip_ad_with_missing_car_type_uses_unknown(settings: Settings) ->
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )
 
@@ -405,7 +405,7 @@ async def test_trip_ad_with_missing_name_uses_none(settings: Settings) -> None:
     service = GroupMessageService(
         repository=repo,
         embeddings=embeddings,
-        provider=provider,
+        ai=provider,
         settings=settings,
     )