Enable IRIS to use FAQs

app/common/PipelineEnum.py

@@ -14,4 +14,6 @@ class PipelineEnum(str, Enum):
     IRIS_SUMMARY_PIPELINE = "IRIS_SUMMARY_PIPELINE"
     IRIS_LECTURE_RETRIEVAL_PIPELINE = "IRIS_LECTURE_RETRIEVAL_PIPELINE"
     IRIS_LECTURE_INGESTION = "IRIS_LECTURE_INGESTION"
+    IRIS_FAQ_INGESTION = "IRIS_FAQ_INGESTION"
+    IRIS_FAQ_RETRIEVAL_PIPELINE = "IRIS_FAQ_RETRIEVAL_PIPELINE"
     NOT_SET = "NOT_SET"

app/domain/data/faq_dto.py

@@ -0,0 +1,10 @@
+from pydantic import BaseModel, Field
+
+
+class FaqDTO(BaseModel):
+    faq_id: int = Field(alias="faqId")
+    course_id: int = Field(alias="courseId")
+    question_title: str = Field(alias="questionTitle")
+    question_answer: str = Field(alias="questionAnswer")
+    course_name: str = Field(default="", alias="courseName")
+    course_description: str = Field(default="", alias="courseDescription")

app/domain/ingestion/deletionPipelineExecutionDto.py

@@ -3,6 +3,7 @@
 from pydantic import Field
 
 from app.domain import PipelineExecutionDTO, PipelineExecutionSettingsDTO
+from app.domain.data.faq_dto import FaqDTO
 from app.domain.data.lecture_unit_dto import LectureUnitDTO
 from app.domain.status.stage_dto import StageDTO
 
@@ -13,3 +14,11 @@ class LecturesDeletionExecutionDto(PipelineExecutionDTO):
     initial_stages: Optional[List[StageDTO]] = Field(
         default=None, alias="initialStages"
     )
+
+
+class FaqDeletionExecutionDto(PipelineExecutionDTO):
+    faq: FaqDTO = Field(..., alias="pyrisFaqWebhookDTO")
+    settings: Optional[PipelineExecutionSettingsDTO]
+    initial_stages: Optional[List[StageDTO]] = Field(
+        default=None, alias="initialStages"
+    )

app/domain/ingestion/ingestion_pipeline_execution_dto.py

@@ -3,6 +3,7 @@
 from pydantic import Field
 
 from app.domain import PipelineExecutionDTO, PipelineExecutionSettingsDTO
+from app.domain.data.faq_dto import FaqDTO
 from app.domain.data.lecture_unit_dto import LectureUnitDTO
 from app.domain.status.stage_dto import StageDTO
 
@@ -13,3 +14,11 @@ class IngestionPipelineExecutionDto(PipelineExecutionDTO):
     initial_stages: Optional[List[StageDTO]] = Field(
         default=None, alias="initialStages"
     )
+
+
+class FaqIngestionPipelineExecutionDto(PipelineExecutionDTO):
+    faq: FaqDTO = Field(..., alias="pyrisFaqWebhookDTO")
+    settings: Optional[PipelineExecutionSettingsDTO]
+    initial_stages: Optional[List[StageDTO]] = Field(
+        default=None, alias="initialStages"
+    )

app/pipeline/chat/course_chat_pipeline.py

@@ -20,7 +20,7 @@
     InteractionSuggestionPipeline,
 )
 from .lecture_chat_pipeline import LectureChatPipeline
-from ..shared.citation_pipeline import CitationPipeline
+from ..shared.citation_pipeline import CitationPipeline, InformationType
 from ..shared.utils import generate_structured_tools_from_functions
 from ...common.message_converters import convert_iris_message_to_langchain_message
 from ...common.pyris_message import PyrisMessage
@@ -42,8 +42,11 @@
 )
 from ...domain import CourseChatPipelineExecutionDTO
 from app.common.PipelineEnum import PipelineEnum
+from ...retrieval.faq_retrieval import FaqRetrieval
+from ...retrieval.faq_retrieval_utils import should_allow_faq_tool, format_faqs
 from ...retrieval.lecture_retrieval import LectureRetrieval
 from ...vector_database.database import VectorDatabase
+from ...vector_database.faq_schema import FaqSchema
 from ...vector_database.lecture_schema import LectureSchema
 from ...web.status.status_update import (
     CourseChatStatusCallback,
@@ -81,6 +84,7 @@ class CourseChatPipeline(Pipeline):
     variant: str
     event: str | None
     retrieved_paragraphs: List[dict] = None
+    retrieved_faqs: List[dict] = None
 
     def __init__(
         self,
@@ -100,19 +104,22 @@ def __init__(
                 requirements=RequirementList(
                     gpt_version_equivalent=4.5,
                 )
-            ), completion_args=completion_args
+            ),
+            completion_args=completion_args,
         )
         self.llm_small = IrisLangchainChatModel(
             request_handler=CapabilityRequestHandler(
                 requirements=RequirementList(
                     gpt_version_equivalent=4.25,
                 )
-            ), completion_args=completion_args
+            ),
+            completion_args=completion_args,
         )
         self.callback = callback
 
         self.db = VectorDatabase()
-        self.retriever = LectureRetrieval(self.db.client)
+        self.lecture_retriever = LectureRetrieval(self.db.client)
+        self.faq_retriever = FaqRetrieval(self.db.client)
         self.suggestion_pipeline = InteractionSuggestionPipeline(variant="course")
         self.citation_pipeline = CitationPipeline()
 
@@ -280,7 +287,7 @@ def lecture_content_retrieval() -> str:
             Only use this once.
             """
             self.callback.in_progress("Retrieving lecture content ...")
-            self.retrieved_paragraphs = self.retriever(
+            self.retrieved_paragraphs = self.lecture_retriever(
                 chat_history=history,
                 student_query=query.contents[0].text_content,
                 result_limit=5,
@@ -300,6 +307,31 @@ def lecture_content_retrieval() -> str:
                 result += lct
             return result
 
+        def faq_content_retrieval() -> str:
+            """
+            Use this tool to retrieve information from indexed FAQs.
+            It is suitable when no other tool fits, it is a common question or the question is frequently asked,
+            or the question could be effectively answered by an FAQ. Also use this if the question is explicitly
+            organizational and course-related. An organizational question about the course might be
+            "What is the course structure?" or "How do I enroll?" or exam related content like "When is the exam".
+            The tool performs a RAG retrieval based on the chat history to find the most relevant FAQs.
+            Each FAQ follows this format: FAQ ID, FAQ Question, FAQ Answer.
+            Respond to the query concisely and solely using the answer from the relevant FAQs.
+            This tool should only be used once per query.
+            """
+            self.callback.in_progress("Retrieving faq content ...")
+            self.retrieved_faqs = self.faq_retriever(
+                chat_history=history,
+                student_query=query.contents[0].text_content,
+                result_limit=10,
+                course_name=dto.course.name,
+                course_id=dto.course.id,
+                base_url=dto.settings.artemis_base_url,
+            )
+
+            result = format_faqs(self.retrieved_faqs)
+            return result
+
         if dto.user.id % 3 < 2:
             iris_initial_system_prompt = tell_iris_initial_system_prompt
             begin_agent_prompt = tell_begin_agent_prompt
@@ -398,6 +430,9 @@ def lecture_content_retrieval() -> str:
             if self.should_allow_lecture_tool(dto.course.id):
                 tool_list.append(lecture_content_retrieval)
 
+            if should_allow_faq_tool(self.db, dto.course.id):
+                tool_list.append(faq_content_retrieval)
+
             tools = generate_structured_tools_from_functions(tool_list)
             # No idea why we need this extra contrary to exercise chat agent in this case, but solves the issue.
             params.update({"tools": tools})
@@ -418,9 +453,19 @@ def lecture_content_retrieval() -> str:
 
             if self.retrieved_paragraphs:
                 self.callback.in_progress("Augmenting response ...")
-                out = self.citation_pipeline(self.retrieved_paragraphs, out)
+                out = self.citation_pipeline(
+                    self.retrieved_paragraphs, out, InformationType.PARAGRAPHS
+                )
             self.tokens.extend(self.citation_pipeline.tokens)
 
+            if self.retrieved_faqs:
+                self.callback.in_progress("Augmenting response ...")
+                out = self.citation_pipeline(
+                    self.retrieved_faqs,
+                    out,
+                    InformationType.FAQS,
+                    base_url=dto.settings.artemis_base_url,
+                )
             self.callback.done("Response created", final_result=out, tokens=self.tokens)
 
             # try:

app/pipeline/chat/exercise_chat_agent_pipeline.py

@@ -39,8 +39,11 @@
 from ...llm import CapabilityRequestHandler, RequirementList
 from ...llm import CompletionArguments
 from ...llm.langchain import IrisLangchainChatModel
+from ...retrieval.faq_retrieval import FaqRetrieval
+from ...retrieval.faq_retrieval_utils import should_allow_faq_tool, format_faqs
 from ...retrieval.lecture_retrieval import LectureRetrieval
 from ...vector_database.database import VectorDatabase
+from ...vector_database.faq_schema import FaqSchema
 from ...vector_database.lecture_schema import LectureSchema
 from weaviate.collections.classes.filters import Filter
 from ...web.status.status_update import ExerciseChatStatusCallback
@@ -103,6 +106,7 @@ class ExerciseChatAgentPipeline(Pipeline):
     prompt: ChatPromptTemplate
     variant: str
     event: str | None
+    retrieved_faqs: List[dict] = None
 
     def __init__(
         self,
@@ -136,7 +140,8 @@ def __init__(
         # Create the pipelines
         self.db = VectorDatabase()
         self.suggestion_pipeline = InteractionSuggestionPipeline(variant="exercise")
-        self.retriever = LectureRetrieval(self.db.client)
+        self.lecture_retriever = LectureRetrieval(self.db.client)
+        self.faq_retriever = FaqRetrieval(self.db.client)
         self.reranker_pipeline = RerankerPipeline()
         self.code_feedback_pipeline = CodeFeedbackPipeline()
         self.pipeline = self.llm_big | JsonOutputParser()
@@ -373,7 +378,7 @@ def lecture_content_retrieval() -> str:
             Only use this once.
             """
             self.callback.in_progress("Retrieving lecture content ...")
-            self.retrieved_paragraphs = self.retriever(
+            self.retrieved_paragraphs = self.lectureRetriever(
                 chat_history=chat_history,
                 student_query=query.contents[0].text_content,
                 result_limit=5,
@@ -393,6 +398,31 @@ def lecture_content_retrieval() -> str:
                 result += lct
             return result
 
+        def faq_content_retrieval() -> str:
+            """
+            Use this tool to retrieve information from indexed FAQs.
+            It is suitable when no other tool fits, it is a common question or the question is frequently asked,
+            or the question could be effectively answered by an FAQ. Also use this if the question is explicitly
+            organizational and course-related. An organizational question about the course might be
+            "What is the course structure?" or "How do I enroll?" or exam related content like "When is the exam".
+            The tool performs a RAG retrieval based on the chat history to find the most relevant FAQs.
+            Each FAQ follows this format: FAQ ID, FAQ Question, FAQ Answer.
+            Respond to the query concisely and solely using the answer from the relevant FAQs.
+            This tool should only be used once per query.
+            """
+            self.callback.in_progress("Retrieving faq content ...")
+            self.retrieved_faqs = self.faq_retriever(
+                chat_history=chat_history,
+                student_query=query.contents[0].text_content,
+                result_limit=10,
+                course_name=dto.course.name,
+                course_id=dto.course.id,
+                base_url=dto.settings.artemis_base_url,
+            )
+
+            result = format_faqs(self.retrieved_faqs)
+            return result
+
         iris_initial_system_prompt = tell_iris_initial_system_prompt
         chat_history_exists_prompt = tell_chat_history_exists_prompt
         no_chat_history_prompt = tell_no_chat_history_prompt
@@ -511,6 +541,10 @@ def lecture_content_retrieval() -> str:
             ]
             if self.should_allow_lecture_tool(dto.course.id):
                 tool_list.append(lecture_content_retrieval)
+
+            if should_allow_faq_tool(self.db, dto.course.id):
+                tool_list.append(faq_content_retrieval)
+
             tools = generate_structured_tools_from_functions(tool_list)
             agent = create_tool_calling_agent(
                 llm=self.llm_big, tools=tools, prompt=self.prompt
@@ -533,7 +567,9 @@ def lecture_content_retrieval() -> str:
                     ]
                 )
 
-                guide_response = (self.prompt | self.llm_small | StrOutputParser()).invoke(
+                guide_response = (
+                    self.prompt | self.llm_small | StrOutputParser()
+                ).invoke(
                     {
                         "response": out,
                     }