Upload 4 files

rag/prompt_template.py

@@ -0,0 +1,95 @@
+from langchain_core.prompts import (
+    ChatPromptTemplate,
+    MessagesPlaceholder,
+    SystemMessagePromptTemplate,
+)
+
+
+def generate_prompt_template():
+    system_template = """`Current date and time: {timestamp}`
+# Role
+
+---
+
+Your name is Ella (Empathetic, Logical, Liaison, Accessible). You are a helpful Virtual Assistant at Tall Tree Health in British Columbia, Canada. Based on the patient's symptoms/needs, connect them with the right practitioner or service offered by Tall Tree. Respond to `Patient Queries` using the `Practitioners Database` and `Tall Tree Health Centre Information` provided in the `Context`. Follow the `Response Guidelines` listed below:
+
+---
+
+# Response Guidelines
+
+1. **Interaction**: Engage in a warm, empathetic, and professional manner. Keep responses brief and focused on the patient's query. Always conclude positively with a reassuring statement. Use markdown formatting and do not use headings.
+
+2. **Symptoms/needs and Location Preference**: Only if not specified, ask for symptoms/needs and location preference (Cordova Bay, James Bay, and Vancouver) before recommending a practitioner or service.
+
+3. **Avoid Making Assumptions**: Stick to the given `Context`. If you're unable to assist, offer the user the contact details for the closest `Tall Tree Health` clinic.
+
+4. Do not give medical advice or act as a health professional. Avoid discussing healthcare costs.
+
+5. **Symptoms/needs and Service Verification**: Match the patient's symptoms/needs with the `Focus Area` field in the `Practitioners Database`. If no match is found, advise the patient accordingly without recommending a practitioner, as Tall Tree is not a primary healthcare provider.
+
+6. **Recommending Practitioners**: Based on the patient's symptoms/needs, location and preferred discipline, recommend only up to 3 practitioners who strictly match the given criteria. Provide the contact info for the corresponding `Tall Tree Health` location for additional assistance.
+
+7. **Practitioner's Contact Information**: Provide contact information in the following structured format. Do not print their `Focus Areas`:
+
+    - `FirstName` and `LastName`:
+    - `Discipline`
+    - [Book an appointment](`BookingLink`) (print only if available)
+
+## Tall Tree Health Service Routing Guidelines
+
+8. **Mental Health Urgent Queries**: For urgent situations such as self-harm, suicidal thoughts, violence, hallucinations, or dissociation direct the patient to call the [9-8-8](tel:9-8-8) suicide crisis helpline, reach out to the Vancouver Island Crisis Line at [1-888-494-3888](tel:1-888-494-3888), or head to the nearest emergency room. Tall Tree isn't equipped for mental health emergencies.
+
+9. **Injuries and Pain**: Prioritize Physiotherapy for injuries and pain conditions unless another preference is stated.
+
+10. **Concussion Protocol**: Direct to the `Concussion Treatment Program` for the appropriate location for a comprehensive assessment with a physiotherapist. Do not recommend a practitioner.
+
+11. **Psychologist in Vancouver**: If a Psychologist is requested in the Vancouver location, provide only the contact and booking link for our mental health team in Cordova Bay - Upstairs location. Do not recommend an alternative practitioner.
+
+12. **Sleep issues**: Recommend only the Sleep Program intake and provide the phone number to book an appointment. Do not recommend a practitioner.
+
+13. **Longevity Program**: For longevity queries, provide the Longevity Program phone number. Do not recommend a practitioner.
+
+14. **DEXA Testing or body composition**: Inform that this service is exclusive to the Cordova Bay clinic and provide the clinic phone number and booking link. Do not recommend a practitioner.
+
+15. **For VO2 Max Testing**: Determine the patient's location preference for Vancouver or Victoria and provide the booking link for the appropriate location. If Victoria, we only do it at our Cordova Bay location.
+
+---
+
+# Patient Query
+
+```
+{message}
+```
+---
+
+# Context
+
+---
+1. **Practitioners Database**:
+
+```
+{practitioners_db}
+```
+---
+
+2. **Tall Tree Health Centre Information**:
+
+```
+{tall_tree_db}
+```
+---
+
+"""
+
+    # Template for system message with markdown formatting
+    system_message = SystemMessagePromptTemplate.from_template(system_template)
+
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            system_message,
+            MessagesPlaceholder(variable_name="history"),
+            ("human", "{message}"),
+        ]
+    )
+
+    return prompt

rag/retrievers.py

@@ -0,0 +1,72 @@
+import os
+from typing import List, Literal
+
+from langchain_core.vectorstores import VectorStoreRetriever
+from langchain_openai import OpenAIEmbeddings
+from langchain_qdrant import FastEmbedSparse, QdrantVectorStore, RetrievalMode
+
+os.environ["GRPC_VERBOSITY"] = "NONE"
+
+
+class RetrieversConfig:
+    def __init__(
+        self,
+        dense_model_name: Literal["text-embedding-3-small"] = "text-embedding-3-small",
+        sparse_model_name: Literal[
+            "prithivida/Splade_PP_en_v1"
+        ] = "prithivida/Splade_PP_en_v1",
+    ):
+        self.required_env_vars = ["QDRANT_API_KEY", "QDRANT_URL", "OPENAI_API_KEY"]
+        self._validate_environment(self.required_env_vars)
+        self.qdrant_url = os.getenv("QDRANT_URL")
+        self.qdrant_api_key = os.getenv("QDRANT_API_KEY")
+        self.dense_embeddings = OpenAIEmbeddings(model=dense_model_name)
+        self.sparse_embeddings = FastEmbedSparse(
+            model_name=sparse_model_name,
+        )
+
+    def _validate_environment(self, required_env_vars: List[str]):
+        missing_vars = [
+            var for var in required_env_vars if not os.getenv(var, "").strip()
+        ]
+        if missing_vars:
+            raise EnvironmentError(
+                f"Missing or empty environment variable(s): {', '.join(missing_vars)}"
+            )
+
+    def get_qdrant_retriever(
+        self,
+        collection_name: str,
+        dense_vector_name: str,
+        sparse_vector_name: str,
+        k: int = 5,
+    ) -> VectorStoreRetriever:
+        qdrantdb = QdrantVectorStore.from_existing_collection(
+            embedding=self.dense_embeddings,
+            sparse_embedding=self.sparse_embeddings,
+            url=self.qdrant_url,
+            api_key=self.qdrant_api_key,
+            prefer_grpc=True,
+            collection_name=collection_name,
+            retrieval_mode=RetrievalMode.HYBRID,
+            vector_name=dense_vector_name,
+            sparse_vector_name=sparse_vector_name,
+        )
+
+        return qdrantdb.as_retriever(search_kwargs={"k": k})
+
+    def get_documents_retriever(self, k: int = 5) -> VectorStoreRetriever:
+        return self.get_qdrant_retriever(
+            collection_name="docs_hybrid_db",
+            dense_vector_name="docs_dense_vectors",
+            sparse_vector_name="docs_sparse_vectors",
+            k=k,
+        )
+
+    def get_practitioners_retriever(self, k: int = 5) -> VectorStoreRetriever:
+        return self.get_qdrant_retriever(
+            collection_name="practitioners_hybrid_db",
+            dense_vector_name="practitioners_dense_vectors",
+            sparse_vector_name="practitioners_sparse_vectors",
+            k=k,
+        )

rag/runnable.py

@@ -0,0 +1,129 @@
+import os
+import random
+from datetime import datetime
+from operator import itemgetter
+from typing import Sequence
+
+import langsmith
+from langchain.memory import ConversationBufferWindowMemory
+from langchain_community.document_transformers import LongContextReorder
+from langchain_core.documents import Document
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.runnables import Runnable, RunnableLambda
+from langchain_openai import ChatOpenAI
+from zoneinfo import ZoneInfo
+
+from rag.retrievers import RetrieversConfig
+
+from .prompt_template import generate_prompt_template
+
+# Helpers
+
+
+def get_datetime() -> str:
+    """Get the current date and time."""
+    return datetime.now(ZoneInfo("America/Vancouver")).strftime("%A, %Y-%b-%d %H:%M:%S")
+
+
+def reorder_documents(docs: list[Document]) -> Sequence[Document]:
+    """Reorder documents to mitigate performance degradation with long contexts."""
+
+    return LongContextReorder().transform_documents(docs)
+
+
+def randomize_documents(documents: list[Document]) -> list[Document]:
+    """Randomize documents to vary model recommendations."""
+    random.shuffle(documents)
+    return documents
+
+
+class DocumentFormatter:
+    def __init__(self, prefix: str):
+        self.prefix = prefix
+
+    def __call__(self, docs: list[Document]) -> str:
+        """Format the Documents to markdown.
+        Args:
+            docs (list[Documents]): List of Langchain documents
+        Returns:
+            docs (str):
+        """
+        return "\n---\n".join(
+            [
+                f"- {self.prefix} {i+1}:\n\n\t" + d.page_content
+                for i, d in enumerate(docs)
+            ]
+        )
+
+
+def create_langsmith_client():
+    """Create a Langsmith client."""
+    os.environ["LANGCHAIN_TRACING_V2"] = "true"
+    os.environ["LANGCHAIN_PROJECT"] = "admin-ai-assistant"
+    os.environ["LANGCHAIN_ENDPOINT"] = "https://api.smith.langchain.com"
+    langsmith_api_key = os.getenv("LANGCHAIN_API_KEY")
+    if not langsmith_api_key:
+        raise EnvironmentError("Missing environment variable: LANGCHAIN_API_KEY")
+    return langsmith.Client()
+
+
+# Set up Runnable and Memory
+
+
+def get_runnable(
+    model: str = "gpt-4o-mini", temperature: float = 0.1
+) -> tuple[Runnable, ConversationBufferWindowMemory]:
+    """Set up runnable and chat memory
+
+    Args:
+        model_name (str, optional): LLM model. Defaults to "gpt-4o".
+        temperature (float, optional): Model temperature. Defaults to 0.1.
+
+    Returns:
+        Runnable, Memory: Chain and Memory
+    """
+
+    # Set up Langsmith to trace the chain
+    create_langsmith_client()
+
+    # LLM and prompt template
+    llm = ChatOpenAI(
+        model=model,
+        temperature=temperature,
+    )
+
+    prompt = generate_prompt_template()
+
+    # Set retrievers with Hybrid search
+
+    retrievers_config = RetrieversConfig()
+
+    # Practitioners data
+    practitioners_data_retriever = retrievers_config.get_practitioners_retriever(k=10)
+
+    # Tall Tree documents with contact information for locations and services
+    documents_retriever = retrievers_config.get_documents_retriever(k=10)
+
+    # Set conversation history window memory. It only uses the last k interactions
+    memory = ConversationBufferWindowMemory(
+        memory_key="history",
+        return_messages=True,
+        k=6,
+    )
+
+    # Set up runnable using LCEL
+    setup = {
+        "practitioners_db": itemgetter("message")
+        | practitioners_data_retriever
+        | DocumentFormatter("Practitioner #"),
+        "tall_tree_db": itemgetter("message")
+        | documents_retriever
+        | DocumentFormatter("No."),
+        "timestamp": lambda _: get_datetime(),
+        "history": RunnableLambda(memory.load_memory_variables) | itemgetter("history"),
+        "message": itemgetter("message"),
+    }
+
+    chain = setup | prompt | llm | StrOutputParser()
+
+    return chain, memory