cleaned up gen ai services

services/py-help-service/main.py

@@ -7,15 +7,21 @@
 from fastapi import Depends, FastAPI, HTTPException, Header, Request
 from dotenv import load_dotenv
 from fastapi.responses import JSONResponse
-from langchain.chat_models import init_chat_model
+from langchain.chat_models import BaseChatModel, init_chat_model
 from langchain_core.messages import HumanMessage, SystemMessage
+from pydantic import BaseModel
+
+from client.cooking_assistant_gen_ai_services_api_internal_client.types import Unset
 
 from client.cooking_assistant_gen_ai_services_api_internal_client.models.help_request_forwarded import (
 	HelpRequestForwarded,
 )
-from client.cooking_assistant_gen_ai_services_api_internal_client.models.help_response import (
-	HelpResponse,
-)
+
+
+# autogenerated classes must be mapped to local ones that are compatible with pydantic
+class LocalHelpResponse(BaseModel):
+	response: str
+
 
 # Load variables from .env for local testing
 load_dotenv()
@@ -118,7 +124,11 @@ def get_llm():
 			raise RuntimeError("CRITICAL: GEMINI_HELP_SERVICE_KEY is missing!")
 
 		kwargs["google_api_key"] = gemini_key
-		kwargs["response_format"] = {"type": "application/json"}
+		kwargs["thinking_level"] = (
+			"low"  # inference time fluctuate a lot
+			# minimal doesn't seem a lot faster than low but seemingly produces more mistakes in json output formatting
+			# medium and high seem noticeably slower than low
+		)
 		model_name = os.getenv("GEMINI_MODEL", "gemini-3.1-flash-lite")
 
 	try:
@@ -133,7 +143,9 @@ def health_check():
 
 
 @app.post("/ai/help", dependencies=[Depends(verify_internal_hmac)])
-async def generate_help(request_data: dict[str, Any], llm=Depends(get_llm)):
+async def generate_help(
+	request_data: dict[str, Any], llm: BaseChatModel = Depends(get_llm)
+):
 	try:
 		request = HelpRequestForwarded.from_dict(request_data)
 	except Exception as e:
@@ -170,20 +182,47 @@ async def generate_help(request_data: dict[str, Any], llm=Depends(get_llm)):
 				)
 
 		if request.recipe:
-			context.append(
+			recipe_ctx = [
 				f"The user is currently looking at a recipe for '{request.recipe.title}'."
+			]
+
+			ingredients = getattr(request.recipe, "ingredients", None)
+			if ingredients:
+				recipe_ctx.append("\nIngredients:")
+				for ing in ingredients:
+					recipe_ctx.append(f"- {ing}")
+
+			instructions = getattr(request.recipe, "instructions", None) or getattr(
+				request.recipe, "steps", None
 			)
+			if instructions:
+				recipe_ctx.append("\nInstructions:")
+				for idx, step in enumerate(instructions, 1):
+					recipe_ctx.append(f"{idx}. {step}")
+
+			if request.recipe.nutrients and not isinstance(
+				request.recipe.nutrients, Unset
+			):
+				nut = request.recipe.nutrients
+				recipe_ctx.append("\nNutritional Information (Total recipe):")
+				recipe_ctx.append(f"- Calories: {nut.calories} kcal")
+				recipe_ctx.append(f"- Protein: {nut.protein}g")
+				recipe_ctx.append(f"- Fat: {nut.fat}g")
+				recipe_ctx.append(f"- Carbohydrates: {nut.carbs}g")
+
+			context.append("\n".join(recipe_ctx))
 
 		# Combine into LLM prompt
 		system_prompt = SystemMessage(content=" ".join(context))
 		user_prompt = HumanMessage(content=request.prompt)
 
-		result = await asyncio.wait_for(
-			llm.ainvoke([system_prompt, user_prompt]), timeout=60
+		structured_llm = llm.with_structured_output(LocalHelpResponse)
+
+		result: LocalHelpResponse = await asyncio.wait_for(
+			structured_llm.ainvoke([system_prompt, user_prompt]), timeout=60
 		)
 
-		help_response = HelpResponse(response=result.content)
-		return help_response.to_dict()
+		return result.model_dump()
 
 	except asyncio.TimeoutError:
 		raise HTTPException(

services/py-help-service/requirements.txt

@@ -1,10 +1,10 @@
 fastapi==0.111.0
 uvicorn==0.30.1
 langchain==1.3.11
-langchain-openai==1.3.3
-langchain-google-genai==2.1.12
 langchain-core==1.4.7
-pydantic==2.7.4
+langchain-openai==1.3.3
+langchain-google-genai==4.2.6
+pydantic==2.12.5
 python-dotenv==1.0.1
 attrs==23.2.0
 python-dateutil==2.9.0.post0

services/py-help-service/tests/test_help_service.py

@@ -3,118 +3,127 @@
 import hmac
 import hashlib
 import json
-from unittest.mock import AsyncMock
+from unittest.mock import AsyncMock, MagicMock
 from fastapi.testclient import TestClient
-from main import app, get_llm, SECRET_KEY_STR
+from main import app, get_llm, SECRET_KEY_STR, LocalHelpResponse
 
 client = TestClient(app)
 
+
 def create_auth_headers(payload: dict):
-    """Helper to generate valid HMAC headers for testing."""
-    timestamp = str(int(time.time()))
-    body_bytes = json.dumps(payload, separators=(',', ':')).encode('utf-8')
-
-    hmac_context = hmac.new(SECRET_KEY_STR.encode('utf-8'), digestmod=hashlib.sha256)
-    hmac_context.update(timestamp.encode('utf-8'))
-    hmac_context.update(b'.')
-    hmac_context.update(body_bytes)
-
-    return {
-        "X-Internal-Timestamp": timestamp,
-        "X-Internal-Signature": hmac_context.hexdigest()
-    }
+	"""Helper to generate valid HMAC headers for testing."""
+	timestamp = str(int(time.time()))
+	body_bytes = json.dumps(payload, separators=(",", ":")).encode("utf-8")
+
+	hmac_context = hmac.new(SECRET_KEY_STR.encode("utf-8"), digestmod=hashlib.sha256)
+	hmac_context.update(timestamp.encode("utf-8"))
+	hmac_context.update(b".")
+	hmac_context.update(body_bytes)
+
+	return {
+		"X-Internal-Timestamp": timestamp,
+		"X-Internal-Signature": hmac_context.hexdigest(),
+	}
+
 
 def test_health_check():
-    response = client.get("/health")
-    assert response.status_code == 200
-    assert response.json() == {"status": "healthy"}
+	response = client.get("/health")
+	assert response.status_code == 200
+	assert response.json() == {"status": "healthy"}
 
 
 @pytest.fixture
 def mock_llm():
-    mock = AsyncMock()
-    app.dependency_overrides[get_llm] = lambda: mock
-    yield mock
-    app.dependency_overrides.clear()
+	mock_base = MagicMock()
+	mock_structured_runnable = AsyncMock()
+
+	# llm.with_structured_output(LocalHelpResponse) returns the runnable
+	mock_base.with_structured_output.return_value = mock_structured_runnable
+	app.dependency_overrides[get_llm] = lambda: mock_base
+	yield mock_base
+	app.dependency_overrides.clear()
 
 
 def test_generate_help_success(mock_llm):
-    mock_llm.ainvoke.return_value.content = "Add a pinch of salt."
-
-    payload = {
-        "profile": {
-            "username": "testuser",
-            "preferences": {
-                "diet": ["vegan"],
-                "allergies": [],
-                "about_me": [],
-                "language": "EN"
-            }
-        },
-        "recipe": {
-            "title": "Tomato Soup",
-            "ingredients": [{"quantity": 1, "unit": "cup", "name": "Tomato"}],
-            "instructions": ["Boil tomatoes."],
-            "portions": 2.0
-        },
-        "prompt": "How do I fix bland soup?"
-    }
-
-    headers = create_auth_headers(payload)
-    response = client.post("/ai/help", json=payload, headers=headers)
-
-    assert response.status_code == 200
-    assert response.json()["response"] == "Add a pinch of salt."
-    mock_llm.ainvoke.assert_called_once()
+	mock_structured_runnable = mock_llm.with_structured_output.return_value
+	mock_structured_runnable.ainvoke.return_value = LocalHelpResponse(
+		response="Add a pinch of salt."
+	)
+
+	payload = {
+		"profile": {
+			"username": "testuser",
+			"preferences": {
+				"diet": ["vegan"],
+				"allergies": [],
+				"about_me": [],
+				"language": "EN",
+			},
+		},
+		"recipe": {
+			"title": "Tomato Soup",
+			"ingredients": [{"quantity": 1, "unit": "cup", "name": "Tomato"}],
+			"instructions": ["Boil tomatoes."],
+			"portions": 2.0,
+		},
+		"prompt": "How do I fix bland soup?",
+	}
+
+	headers = create_auth_headers(payload)
+	response = client.post("/ai/help", json=payload, headers=headers)
+
+	assert response.status_code == 200
+	assert response.json()["response"] == "Add a pinch of salt."
+
+	mock_structured_runnable.ainvoke.assert_called_once()
+
 
 def test_generate_help_invalid_payload(mock_llm):
-
-    # preferences field missing
-    payload = {
-        "profile": {
-            "username": "testuser"
-        },
-        "recipe": {
-            "title": "Tomato Soup",
-            "ingredients": [{"quantity": 1, "unit": "cup", "name": "Tomato"}],
-            "instructions": ["Boil tomatoes."],
-            "portions": 2.0
-        },
-        "prompt": "How do I fix bland soup?"
-    }
-
-    headers = create_auth_headers(payload)
-    response = client.post("/ai/help", json=payload, headers=headers)
-
-    assert response.status_code == 400
-
-    response_json = response.json()
-    if "detail" in response_json:
-        assert "Invalid request format" in str(response_json["detail"])
-    else:
-        assert "Invalid request format" in response_json["message"]
+
+	# preferences field missing
+	payload = {
+		"profile": {"username": "testuser"},
+		"recipe": {
+			"title": "Tomato Soup",
+			"ingredients": [{"quantity": 1, "unit": "cup", "name": "Tomato"}],
+			"instructions": ["Boil tomatoes."],
+			"portions": 2.0,
+		},
+		"prompt": "How do I fix bland soup?",
+	}
+
+	headers = create_auth_headers(payload)
+	response = client.post("/ai/help", json=payload, headers=headers)
+
+	assert response.status_code == 400
+
+	response_json = response.json()
+	if "detail" in response_json:
+		assert "Invalid request format" in str(response_json["detail"])
+	else:
+		assert "Invalid request format" in response_json["message"]
 
 
 def test_generate_help_unauthorized():
-    # no header
-    response = client.post("/ai/help", json={"prompt": "test"})
-    assert response.status_code == 401
+	# no header
+	response = client.post("/ai/help", json={"prompt": "test"})
+	assert response.status_code == 401
 
 
 def test_generate_help_invalid_signature():
-    payload = {"prompt": "test"}
-    headers = {
-        "X-Internal-Timestamp": str(int(time.time())),
-        "X-Internal-Signature": "wrong-signature"
-    }
-
-    response = client.post("/ai/help", json=payload, headers=headers)
-    assert response.status_code == 403
-    
-    response_json = response.json()
-    if "detail" in response_json:
-        assert "message" in response_json["detail"]
-        assert "Forbidden" in response_json["detail"]["message"]
-    else:
-        assert "message" in response_json
-        assert "Forbidden" in response_json["message"]
+	payload = {"prompt": "test"}
+	headers = {
+		"X-Internal-Timestamp": str(int(time.time())),
+		"X-Internal-Signature": "wrong-signature",
+	}
+
+	response = client.post("/ai/help", json=payload, headers=headers)
+	assert response.status_code == 403
+
+	response_json = response.json()
+	if "detail" in response_json:
+		assert "message" in response_json["detail"]
+		assert "Forbidden" in response_json["detail"]["message"]
+	else:
+		assert "message" in response_json
+		assert "Forbidden" in response_json["message"]