initial pieces for llm-based question classification

kitsune/llm/__init__.py

@@ -0,0 +1,15 @@
+from functools import cache
+
+from langchain_google_vertexai import ChatVertexAI
+
+
+@cache
+def get_llm(
+    model_name: str, temperature: int = 0, max_tokens: int | None = None, max_retries: int = 2
+) -> ChatVertexAI:
+    """
+    Returns a LangChain chat model instance based on the given LLM model name.
+    """
+    return ChatVertexAI(
+        model=model_name, temperature=temperature, max_tokens=max_tokens, max_retries=max_retries
+    )

kitsune/llm/apps.py

@@ -0,0 +1,6 @@
+from django.apps import AppConfig
+
+
+class LLMConfig(AppConfig):
+    name = "kitsune.llm"
+    default_auto_field = "django.db.models.AutoField"

kitsune/llm/questions/prompt.py

@@ -0,0 +1,98 @@
+from langchain.output_parsers import ResponseSchema, StructuredOutputParser
+from langchain.prompts import ChatPromptTemplate
+
+
+SPAM_INSTRUCTIONS = """
+# Role and goal
+- You are a moderation agent, and an expert on Mozilla's "{product}" product.
+- Your goal is to determine if the given question is spam.
+- The question should be relevant to Mozilla's "{product}" product, but it may be spam.
+
+# Spam cases
+In each of the following cases, the question should be considered spam:
+  - It is trying to sell, advertise, or promote a product or service.
+  - It encourages users to call a phone number or contact an individual/business.
+  - It contains coupons, coupon codes, or discount codes.
+  - It contains or links to sexually explicit or pornographic content.
+  - It contains or promotes hateful, violent, racist, or abusive speech.
+  - It encourages illegal, unethical, or dangerous activities.
+  - It is trying to promote a political point of view.
+  - It is a single word or very short with no clear intent.
+  - It is impossible to determine the intent or purpose of the question.
+  - It contains an excessive amount of random symbols/emojis.
+  - It contains QR codes or images containing external links.
+  - It is not relevant to Mozilla's "{product}" product.
+
+# Specific instructions
+Given a question, you must do the following:
+- Determine whether or not the question is spam by considering each of the spam cases
+  listed above.
+- Provide a reason for your determination.
+
+# Response format instructions
+{format_instructions}
+"""
+
+TOPIC_INSTRUCTIONS = """
+# Role and goal
+- You are a classification agent, and an expert on Mozilla's "{product}" product.
+- Your goal is to select the best topic for classifying a question about Mozilla's
+  "{product}" product.
+
+# Eligible Topics
+Below are the hierarchical topics you MUST select from, provided in JSON format. Each
+topic includes a title and a description, and may also include some examples that will
+help you identify the kinds of questions which should be classified under that topic:
+
+```json
+{topics}
+```
+
+# Specific instructions
+- Given a question, consider the title, description, and examples of each one of the
+  eligible topics listed above, and then select the most relevant topic.
+- If you conclude that the question does not relate to Mozilla's "{product}" product,
+  or does not provide enough information to make a selection, select the "Undefined" topic.
+- Provide a reason for your selection.
+
+# Response format instructions
+{format_instructions}
+"""
+
+
+spam_parser = StructuredOutputParser.from_response_schemas(
+    (
+        ResponseSchema(
+            name="is_spam",
+            type="bool",
+            description="A boolean that when true indicates that the question is spam.",
+        ),
+        ResponseSchema(
+            name="reason",
+            type="string",
+            description="The reason for identifying the question as spam or not spam.",
+        ),
+    )
+)
+
+
+topic_parser = StructuredOutputParser.from_response_schemas(
+    (
+        ResponseSchema(
+            name="topic",
+            type="string",
+            description="The topic selected for the question.",
+        ),
+        ResponseSchema(
+            name="reason",
+            type="string",
+            description="The reason for selecting the topic.",
+        ),
+    )
+)
+
+
+spam_prompt = ChatPromptTemplate((("system", SPAM_INSTRUCTIONS), ("human", "{question}")))
+
+
+topic_prompt = ChatPromptTemplate((("system", TOPIC_INSTRUCTIONS), ("human", "{question}")))