First end-to-end working version

2024-02-13 23:47:10 +00:00 · 2024-02-13 23:47:10 +00:00 · abca2365bf
parent 89e52cbdb1
commit abca2365bf
25 changed files with 809 additions and 75 deletions
--- a/.devcontainer/compose.yaml
+++ b/.devcontainer/compose.yaml
@ -8,9 +8,12 @@ services:
    ports:
      - "8000:8000"
    environment:
-      - QDRANT_URL=qdrant:6334
+      - QDRANT_HOST=qdrant
+      - QQDRANT_GRPC_PORT=6334
      - TEI_BASE_URL=http://text-embeddings-inference
      - TEI_RERANK_BASE_URL=http://text-embeddings-inference-rerank
+      - OLLAMA_BASE_URL=http://ollama:11434
+      - OLLAMA_MODEL_NAME=${OLLAMA_MODEL_NAME}
    command: sleep infinity

  qdrant:
@ -30,6 +33,7 @@ services:
    environment:
      - MODEL_ID=${TEI_MODEL_ID:-BAAI/bge-large-en-v1.5}
      - REVISION=${TEI_MODEL_REVISION}
+      - MAX_CLIENT_BATCH_SIZE=128

  text-embeddings-inference-rerank:
    image: ghcr.io/huggingface/text-embeddings-inference:cpu-0.6
@ -40,3 +44,19 @@ services:
    environment:
      - MODEL_ID=${TEI_RERANK_MODEL_ID:-BAAI/bge-reranker-large}
      - REVISION=${TEI_RERANK_MODEL_REVISION}
+      - MAX_CLIENT_BATCH_SIZE=128
+
+  ollama:
+    image: ollama/ollama:latest
+    volumes:
+      - ../ollama:/root/.ollama
+    ports:
+      - 5000:11434
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: ${OLLAMA_GPU_DRIVER-nvidia}
+              count: ${OLLAMA_GPU_COUNT-1}
+              capabilities:
+                - gpu
--- a/.gitignore
+++ b/.gitignore
@ -163,3 +163,4 @@ cython_debug/
 data/
 tei_data/
 qdrant_storage/
+ollama/
--- a/llm-qa/llm_qa/chains/text_splitters/markdown_header_text_splitter.py
+++ b/llm-qa/llm_qa/chains/text_splitters/markdown_header_text_splitter.py
@ -6,7 +6,7 @@ def get_markdown_header_text_splitter_chain(
    markdown_header_text_splitter: MarkdownHeaderTextSplitter,
 ) -> Runnable[str, list[str]]:
    if not markdown_header_text_splitter.strip_headers:
-        raise ValueError("`strip_headers` must be True")  # noqa: TRY003
+        raise ValueError("`strip_headers` must be True")

    @chain
    def markdown_header_text_splitter_chain(text: str) -> list[str]:
@ -15,7 +15,11 @@ def get_markdown_header_text_splitter_chain(
        return [
            "\n...\n".join(
                f"{header_key} {document.metadata[header_key]}"
-                for _, header_key in markdown_header_text_splitter.headers_to_split_on
+                for _, header_key in sorted(
+                    markdown_header_text_splitter.headers_to_split_on,
+                    key=lambda x: len(x[0]),
+                )
+                if header_key in document.metadata
            )
            + f"\n{document.page_content}"
            for document in documents
--- a/llm-qa/llm_qa/client.py
+++ b/llm-qa/llm_qa/client.py
@ -0,0 +1,114 @@
+import logging
+import pathlib
+from typing import Annotated, Optional
+
+import httpx
+import typer
+
+from llm_qa.logging import load_logging_config
+from llm_qa.models.chat import ChatMessage, ChatRequest, ChatResponse
+from llm_qa.models.prompts import ChatPrompts
+from llm_qa.models.upsert import TextType, UpsertTextRequest, UpsertTextResponse
+
+logger = logging.getLogger(__name__)
+
+SYSTEM_MESSAGE_PROMPT_TEMPLATE_PATH = "../data/prompts/system_message.txt"
+LAST_HUMAN_MESSAGE_PROMPT_TEMPLATE_PATH = "../data/prompts/last_human_message.txt"
+CHAT_URL = "http://localhost:8000/api/v1/chat"
+UPSERT_URL = "http://localhost:8000/api/v1/upsert-text"
+
+
+def load_prompt(path: str) -> str:
+    return pathlib.Path(path).read_text()
+
+
+app = typer.Typer()
+
+
+@app.command()
+def chat(
+    collection: Annotated[str, typer.Option()],
+    chat_url: Annotated[str, typer.Option()] = CHAT_URL,
+    system_message_prompt_template_path: Annotated[
+        Optional[typer.FileText], typer.Option()  # noqa: UP007
+    ] = None,
+    last_human_message_prompt_template_path: Annotated[
+        Optional[typer.FileText], typer.Option()  # noqa: UP007
+    ] = None,
+) -> None:
+    system_message_prompt_template = (
+        system_message_prompt_template_path.read()
+        if system_message_prompt_template_path
+        else load_prompt(SYSTEM_MESSAGE_PROMPT_TEMPLATE_PATH)
+    )
+    last_human_message_prompt_template = (
+        last_human_message_prompt_template_path.read()
+        if last_human_message_prompt_template_path
+        else load_prompt(LAST_HUMAN_MESSAGE_PROMPT_TEMPLATE_PATH)
+    )
+    prompts = ChatPrompts(
+        system_message=system_message_prompt_template,
+        last_human_message=last_human_message_prompt_template,
+    )
+    client = httpx.Client(timeout=180.0)
+    messages: list[ChatMessage] = []
+    while True:
+        message_content = input("### User: ")
+        if message_content in {"/exit", "/quit", "/bye"}:
+            break
+        message = ChatMessage.new_human(message_content)
+        messages.append(message)
+        response = client.post(
+            chat_url,
+            json=ChatRequest(
+                messages=messages, collection_name=collection, prompts=prompts
+            ).model_dump(),
+        )
+        response.raise_for_status()
+        chat_response = ChatResponse.model_validate_json(response.content)
+        messages.append(chat_response.response_message)
+        print("### AI:", chat_response.response_message.content)
+        print(
+            "----------\n### Sources:\n"
+            + "\n\n".join(
+                f"Source {i + 1}:\n{source.content}"
+                for i, source in enumerate(chat_response.sources)
+            )
+            + "\n----------"
+        )
+
+
+@app.command()
+def upsert(
+    files: list[typer.FileText],
+    collection: Annotated[str, typer.Option()],
+    upsert_url: Annotated[str, typer.Option()] = UPSERT_URL,
+) -> None:
+    client = httpx.Client(timeout=180.0)
+    for file in files:
+        logger = logging.getLogger(__name__)
+        text = file.read()
+        ext = pathlib.Path(file.name).suffix
+        match ext.lower():
+            case ".txt":
+                type = TextType.PLAIN_TEXT
+            case ".md":
+                type = TextType.MARKDOWN
+            case _:
+                logger.error("Unsupported file extension `%s`", ext)
+                continue
+        logger.info("Upserting file `%s` as type `%s`", file.name, type.value)
+        response = client.post(
+            upsert_url,
+            json=UpsertTextRequest(
+                text=text, type=type, collection=collection
+            ).model_dump(),
+        )
+        response.raise_for_status()
+        upsert_response = UpsertTextResponse.model_validate_json(response.content)
+        logger.info("Upserted %d document chunks", upsert_response.num_documents)
+
+
+if __name__ == "__main__":
+    load_logging_config()
+    app()
--- a/llm-qa/llm_qa/dependencies.py
+++ b/llm-qa/llm_qa/dependencies.py
@ -1,8 +1,12 @@
 from typing import Annotated

 from fastapi import Depends
+from langchain.callbacks import StreamingStdOutCallbackHandler
+from langchain.callbacks.manager import CallbackManager
+from langchain.chat_models.base import BaseChatModel
+from langchain.chat_models.ollama import ChatOllama

-from llm_qa.embeddings.tei import TeiEmbeddings
+from llm_qa.models.tei import TeiConfig
 from llm_qa.settings import Settings


@ -10,15 +14,28 @@ def settings() -> Settings:
    return Settings()


-def tei_embeddings(settings: Annotated[Settings, Depends(settings)]) -> TeiEmbeddings:
-    return TeiEmbeddings(
+def tei_config(settings: Annotated[Settings, Depends(settings)]) -> TeiConfig:
+    return TeiConfig(
        base_url=settings.tei_base_url,
+        document_prefix=settings.tei_document_prefix,
+        query_prefix=settings.tei_query_prefix,
    )


-def tei_rerank_embeddings(
+def tei_rerank_config(
    settings: Annotated[Settings, Depends(settings)],
-) -> TeiEmbeddings:
-    return TeiEmbeddings(
+) -> TeiConfig:
+    return TeiConfig(
        base_url=settings.tei_rerank_base_url,
    )
+
+
+def chat_model(
+    settings: Annotated[Settings, Depends(settings)],
+) -> BaseChatModel:
+    return ChatOllama(
+        base_url=settings.ollama_base_url,
+        model=settings.ollama_model_name,
+        verbose=True,
+        callback_manager=CallbackManager([StreamingStdOutCallbackHandler()]),
+    )
--- a/llm-qa/llm_qa/embeddings/base.py
+++ b/llm-qa/llm_qa/embeddings/base.py
@ -1,8 +0,0 @@
-from abc import ABC
-
-from langchain.embeddings.base import Embeddings
-from pydantic import BaseModel
-
-
-class PydanticEmbeddings(Embeddings, BaseModel, ABC):
-    pass
--- a/llm-qa/llm_qa/embeddings/tei.py
+++ b/llm-qa/llm_qa/embeddings/tei.py
@ -2,8 +2,8 @@ from typing import override
 from urllib.parse import urljoin

 import httpx
+from langchain.embeddings.base import Embeddings

-from llm_qa.embeddings.base import PydanticEmbeddings
 from llm_qa.models.tei import (
    EmbedRequest,
    EmbedResponseAdapter,
@ -11,25 +11,28 @@ from llm_qa.models.tei import (
    RerankRequest,
    RerankResponse,
    RerankResponseAdapter,
+    TeiConfig,
 )


-class TeiEmbeddings(PydanticEmbeddings):
-    base_url: str
-    embed_endpoint: str = "/embed"
-    rerank_endpoint: str = "/rerank"
-    document_prefix: str = "passage: "
-    query_prefix: str = "query: "
-    _client: httpx.Client = httpx.Client()
-    _async_client: httpx.AsyncClient = httpx.AsyncClient()
+class TeiEmbeddings(Embeddings):
+    def __init__(
+        self,
+        tei_config: TeiConfig,
+        client: httpx.Client | None = None,
+        async_client: httpx.AsyncClient | None = None,
+    ) -> None:
+        self.tei_config = tei_config
+        self._client = client or httpx.Client(timeout=180.0)
+        self._async_client = async_client or httpx.AsyncClient(timeout=180.0)

    @property
    def embed_url(self) -> str:
-        return urljoin(self.base_url, self.embed_endpoint)
+        return urljoin(self.tei_config.base_url, self.tei_config.embed_endpoint)

    @property
    def rerank_url(self) -> str:
-        return urljoin(self.base_url, self.rerank_endpoint)
+        return urljoin(self.tei_config.base_url, self.tei_config.rerank_endpoint)

    @staticmethod
    def _handle_status(response: httpx.Response) -> None:
@ -46,7 +49,8 @@ class TeiEmbeddings(PydanticEmbeddings):
                    except ValueError:
                        note = e.response.text
                    e.add_note(note)
-            raise
+                case _:
+                    raise

    def _embed(self, text: str | list[str]) -> list[list[float]]:
        """Embed text."""
@ -69,22 +73,24 @@ class TeiEmbeddings(PydanticEmbeddings):
    @override
    def embed_documents(self, texts: list[str]) -> list[list[float]]:
        """Embed search docs."""
-        return self._embed([self.document_prefix + text for text in texts])
+        return self._embed([self.tei_config.document_prefix + text for text in texts])

    @override
    def embed_query(self, text: str) -> list[float]:
        """Embed query text."""
-        return self._embed(self.document_prefix + text)[0]
+        return self._embed(self.tei_config.document_prefix + text)[0]

    @override
    async def aembed_documents(self, texts: list[str]) -> list[list[float]]:
        """Asynchronous Embed search docs."""
-        return await self._aembed([self.document_prefix + text for text in texts])
+        return await self._aembed([
+            self.tei_config.document_prefix + text for text in texts
+        ])

    @override
    async def aembed_query(self, text: str) -> list[float]:
        """Asynchronous Embed query text."""
-        return (await self._aembed(self.document_prefix + text))[0]
+        return (await self._aembed(self.tei_config.document_prefix + text))[0]

    def rerank(self, query: str, texts: list[str]) -> list[RerankResponse]:
        """Rerank texts."""
--- a/llm-qa/llm_qa/logging.py
+++ b/llm-qa/llm_qa/logging.py
@ -0,0 +1,9 @@
+import logging.config
+import pathlib
+
+import yaml
+
+
+def load_logging_config() -> None:
+    logging_config = yaml.safe_load(pathlib.Path("logging.yaml").read_text())
+    logging.config.dictConfig(logging_config)
--- a/llm-qa/llm_qa/models/chat.py
+++ b/llm-qa/llm_qa/models/chat.py
@ -0,0 +1,57 @@
+from enum import StrEnum
+from typing import Self
+
+from langchain_core.messages import AIMessage, BaseMessage, HumanMessage
+from pydantic import BaseModel
+
+from llm_qa.models.prompts import ChatPrompts
+from llm_qa.models.source import Source
+from llm_qa.typing_utils import assert_never
+
+
+class MessageType(StrEnum):
+    HUMAN = "HUMAN"
+    AI = "AI"
+
+
+class ChatMessage(BaseModel):
+    content: str
+    type: MessageType
+
+    @classmethod
+    def new_human(cls, content: str) -> Self:
+        return cls(content=content, type=MessageType.HUMAN)
+
+    @classmethod
+    def new_ai(cls, content: str) -> Self:
+        return cls(content=content, type=MessageType.AI)
+
+    @classmethod
+    def from_langchain_message(cls, message: HumanMessage | AIMessage) -> Self:
+        match message:
+            case HumanMessage(content=content):
+                return cls(content=content, type=MessageType.HUMAN)
+            case AIMessage(content=content):
+                return cls(content=content, type=MessageType.AI)
+            case _:
+                return assert_never(message)
+
+    def to_langchain_message(self) -> BaseMessage:
+        match self.type:
+            case MessageType.HUMAN:
+                return HumanMessage(content=self.content)
+            case MessageType.AI:
+                return AIMessage(content=self.content)
+            case _:
+                return assert_never(self)
+
+
+class ChatRequest(BaseModel):
+    messages: list[ChatMessage]
+    collection_name: str
+    prompts: ChatPrompts | None = None
+
+
+class ChatResponse(BaseModel):
+    response_message: ChatMessage
+    sources: list[Source]
--- a/llm-qa/llm_qa/models/prompts.py
+++ b/llm-qa/llm_qa/models/prompts.py
@ -0,0 +1,6 @@
+from pydantic import BaseModel
+
+
+class ChatPrompts(BaseModel):
+    system_message: str
+    last_human_message: str
--- a/llm-qa/llm_qa/models/source.py
+++ b/llm-qa/llm_qa/models/source.py
@ -0,0 +1,12 @@
+from typing import Self
+
+from langchain_core.documents import Document
+from pydantic import BaseModel
+
+
+class Source(BaseModel):
+    content: str
+
+    @classmethod
+    def from_document(cls, document: Document) -> Self:
+        return cls(content=document.page_content)
--- a/llm-qa/llm_qa/models/tei.py
+++ b/llm-qa/llm_qa/models/tei.py
@ -3,10 +3,18 @@ from enum import StrEnum
 from pydantic import BaseModel, Field, TypeAdapter


+class TeiConfig(BaseModel):
+    base_url: str
+    embed_endpoint: str = "/embed"
+    rerank_endpoint: str = "/rerank"
+    document_prefix: str = ""
+    query_prefix: str = ""
+
+
 class EmbedRequest(BaseModel):
    inputs: str | list[str]
    normalize: bool = True
-    truncate: bool = False
+    truncate: bool = True


 EmbedResponseAdapter = TypeAdapter(list[list[float]])
--- a/llm-qa/llm_qa/routers/api_v1.py
+++ b/llm-qa/llm_qa/routers/api_v1.py
@ -1,6 +1,7 @@
 from fastapi import APIRouter

-from llm_qa.routers import upsert
+from llm_qa.routers import chat, upsert

 router = APIRouter(prefix="/api/v1", tags=["v1"])
 router.include_router(upsert.router)
+router.include_router(chat.router)
--- a/llm-qa/llm_qa/routers/chat.py
+++ b/llm-qa/llm_qa/routers/chat.py
@ -0,0 +1,31 @@
+from typing import Annotated
+
+from fastapi import APIRouter, Depends
+from langchain.chat_models.base import BaseChatModel
+
+from llm_qa.dependencies import chat_model, settings, tei_config
+from llm_qa.models.chat import ChatRequest, ChatResponse
+from llm_qa.models.tei import TeiConfig
+from llm_qa.services.chat import chat as chat_service
+from llm_qa.settings import Settings
+
+router = APIRouter()
+
+
+@router.post("/chat")
+async def upsert_text(
+    chat_request: ChatRequest,
+    settings: Annotated[Settings, Depends(settings)],
+    tei_config: Annotated[TeiConfig, Depends(tei_config)],
+    chat_model: Annotated[BaseChatModel, Depends(chat_model)],
+) -> ChatResponse:
+    prompts = chat_request.prompts or settings.chat_prompts
+    return await chat_service(
+        messages=[message.to_langchain_message() for message in chat_request.messages],
+        collection_name=chat_request.collection_name,
+        prompts=prompts,
+        tei_config=tei_config,
+        qdrant_host=settings.qdrant_host,
+        qdrant_grpc_port=settings.qdrant_grpc_port,
+        chat_model=chat_model,
+    )
--- a/llm-qa/llm_qa/routers/upsert.py
+++ b/llm-qa/llm_qa/routers/upsert.py
@ -2,8 +2,8 @@ from typing import Annotated

 from fastapi import APIRouter, Depends

-from llm_qa.dependencies import settings, tei_embeddings
-from llm_qa.embeddings.base import PydanticEmbeddings
+from llm_qa.dependencies import settings, tei_config
+from llm_qa.models.tei import TeiConfig
 from llm_qa.models.upsert import UpsertTextRequest, UpsertTextResponse
 from llm_qa.services.upsert import upsert_text as upsert_text_service
 from llm_qa.settings import Settings
@ -15,14 +15,15 @@ router = APIRouter()
 async def upsert_text(
    upsert_request: UpsertTextRequest,
    settings: Annotated[Settings, Depends(settings)],
-    embeddings: Annotated[PydanticEmbeddings, Depends(tei_embeddings)],
+    tei_config: Annotated[TeiConfig, Depends(tei_config)],
 ) -> UpsertTextResponse:
    num_documents = await upsert_text_service(
        text=upsert_request.text,
        text_type=upsert_request.type,
        collection=upsert_request.collection,
-        embeddings=embeddings,
-        qdrant_url=settings.qdrant_url,
+        tei_config=tei_config,
+        qdrant_host=settings.qdrant_host,
+        qdrant_grpc_port=settings.qdrant_grpc_port,
    )
    return UpsertTextResponse(num_documents=num_documents)

--- a/llm-qa/llm_qa/services/chat.py
+++ b/llm-qa/llm_qa/services/chat.py
@ -0,0 +1,113 @@
+from operator import itemgetter
+from typing import TypedDict
+
+from langchain.chat_models.base import BaseChatModel
+from langchain.prompts import (
+    ChatPromptTemplate,
+    HumanMessagePromptTemplate,
+    MessagesPlaceholder,
+    SystemMessagePromptTemplate,
+)
+from langchain.vectorstores import VectorStore
+from langchain.vectorstores.qdrant import Qdrant
+from langchain_core.documents import Document
+from langchain_core.messages import AIMessage, BaseMessage
+from langchain_core.runnables import (
+    Runnable,
+    RunnableLambda,
+    RunnablePassthrough,
+    chain,
+)
+from qdrant_client import AsyncQdrantClient, QdrantClient
+
+from llm_qa.embeddings.tei import TeiEmbeddings
+from llm_qa.models.chat import ChatMessage, ChatResponse
+from llm_qa.models.prompts import ChatPrompts
+from llm_qa.models.source import Source
+from llm_qa.models.tei import TeiConfig
+
+
+class ChatRunnableInput(TypedDict):
+    messages: list[BaseMessage]
+    collection_name: str
+
+
+class ChatRunnableOutput(TypedDict):
+    response: AIMessage
+    documents: list[Document]
+
+
+@chain
+def combine_documents(documents: list[Document]) -> str:
+    return "\n\n".join(document.page_content for document in documents)
+
+
+def get_chat_chain(
+    chat_prompt_template: ChatPromptTemplate,
+    vectorstore: VectorStore,
+    chat_model: BaseChatModel,
+) -> Runnable[ChatRunnableInput, ChatRunnableOutput]:
+    return (
+        RunnablePassthrough[ChatRunnableInput]()
+        | {
+            "messages": itemgetter("messages"),
+            "collection_name": itemgetter("collection_name"),
+            "last_human_message": itemgetter("messages")
+            | RunnableLambda[list[BaseMessage], str](lambda x: x[-1].content),
+        }
+        | {
+            "conversation_history": RunnableLambda(lambda x: x["messages"][:-1]),
+            "last_human_message": itemgetter("last_human_message"),
+            "documents": itemgetter("last_human_message") | vectorstore.as_retriever(),
+        }
+        | {
+            "conversation_history": itemgetter("conversation_history"),
+            "last_human_message": itemgetter("last_human_message"),
+            "documents": itemgetter("documents"),
+            "context": itemgetter("documents") | combine_documents,
+        }
+        | {
+            "documents": itemgetter("documents"),
+            "response": chat_prompt_template | chat_model,
+        }
+    )
+
+
+async def chat(
+    messages: list[BaseMessage],
+    collection_name: str,
+    prompts: ChatPrompts,
+    tei_config: TeiConfig,
+    qdrant_host: str,
+    qdrant_grpc_port: int,
+    chat_model: BaseChatModel,
+) -> ChatResponse:
+    chat_prompt_template = ChatPromptTemplate.from_messages([
+        SystemMessagePromptTemplate.from_template(prompts.system_message),
+        MessagesPlaceholder(variable_name="conversation_history"),
+        HumanMessagePromptTemplate.from_template(prompts.last_human_message),
+    ])
+    embeddings = TeiEmbeddings(tei_config=tei_config)
+    qdrant_client = QdrantClient(
+        location=qdrant_host, grpc_port=qdrant_grpc_port, prefer_grpc=True
+    )
+    async_qdrant_client = AsyncQdrantClient(
+        location=qdrant_host, grpc_port=qdrant_grpc_port, prefer_grpc=True
+    )
+    qdrant_vectorstore = Qdrant(
+        client=qdrant_client,
+        async_client=async_qdrant_client,
+        collection_name=collection_name,
+        embeddings=embeddings,
+    )
+    chain = get_chat_chain(chat_prompt_template, qdrant_vectorstore, chat_model)
+    chain_output = await chain.ainvoke(
+        ChatRunnableInput(messages=messages, collection_name=collection_name)
+    )
+    return ChatResponse(
+        response_message=ChatMessage.from_langchain_message(chain_output["response"]),
+        sources=[
+            Source(content=document.page_content)
+            for document in chain_output["documents"]
+        ],
+    )
--- a/llm-qa/llm_qa/services/upsert.py
+++ b/llm-qa/llm_qa/services/upsert.py
@ -1,7 +1,7 @@
 import logging

 from langchain.schema.document import Document
-from langchain_core.embeddings import Embeddings
+from langchain.vectorstores.qdrant import Qdrant

 from llm_qa.chains.text_splitters.markdown_header_text_splitter import (
    markdown_3_headers_text_splitter_chain,
@ -9,8 +9,9 @@ from llm_qa.chains.text_splitters.markdown_header_text_splitter import (
 from llm_qa.chains.text_splitters.text_splitter import (
    recursive_character_text_splitter_chain,
 )
+from llm_qa.embeddings.tei import TeiEmbeddings
+from llm_qa.models.tei import TeiConfig
 from llm_qa.models.upsert import TextType
-from llm_qa.vectorstores.qdrant import upsert_documents

 logger = logging.getLogger(__name__)

@ -19,8 +20,9 @@ async def upsert_text(
    text: str,
    text_type: TextType,
    collection: str,
-    embeddings: Embeddings,
-    qdrant_url: str,
+    tei_config: TeiConfig,
+    qdrant_host: str,
+    qdrant_grpc_port: int,
 ) -> int:
    match text_type:
        case TextType.PLAIN_TEXT:
@ -28,16 +30,25 @@ async def upsert_text(
        case TextType.MARKDOWN:
            text_splitter_chain = markdown_3_headers_text_splitter_chain
        case _:
-            raise ValueError(f"Unknown text type: `{text_type}`")  # noqa: TRY003
+            raise ValueError(f"Unknown text type: `{text_type}`")

    text_chunks = await text_splitter_chain.ainvoke(text)

    documents = [Document(page_content=chunk) for chunk in text_chunks]

-    await upsert_documents(
-        documents=documents,
-        embeddings=embeddings,
-        qdrant_url=qdrant_url,
-        collection=collection,
+    embeddings = TeiEmbeddings(tei_config=tei_config)
+
+    logger.info(
+        "Upserting %d documents to Qdrant collection `%s`", len(documents), collection
    )
+    await Qdrant.afrom_documents(
+        location=qdrant_host,
+        grpc_port=qdrant_grpc_port,
+        prefer_grpc=True,
+        documents=documents,
+        embedding=embeddings,
+        collection_name=collection,
+        force_recreate=False,
+    )
+
    return len(documents)
--- a/llm-qa/llm_qa/settings.py
+++ b/llm-qa/llm_qa/settings.py
@ -1,9 +1,19 @@
 from pydantic_settings import BaseSettings, SettingsConfigDict

+from llm_qa.models.prompts import ChatPrompts
+

 class Settings(BaseSettings):
    model_config = SettingsConfigDict(env_file=".env", env_file_encoding="utf-8")

-    qdrant_url: str
+    qdrant_host: str
+    qdrant_grpc_port: int = 6334
    tei_base_url: str
    tei_rerank_base_url: str
+    tei_document_prefix: str = "passage: "
+    tei_query_prefix: str = "query: "
+    ollama_base_url: str
+    ollama_model_name: str
+    chat_prompts: ChatPrompts = ChatPrompts(
+        system_message="System message", last_human_message="Last human message"
+    )
--- a/llm-qa/llm_qa/typing_utils.py
+++ b/llm-qa/llm_qa/typing_utils.py
@ -0,0 +1,5 @@
+from typing import Never, NoReturn
+
+
+def assert_never(arg: Never) -> NoReturn:
+    raise AssertionError(f"Expected code is unreachable. Instead received `{arg}`.")
--- a/llm-qa/llm_qa/vectorstores/init.py
+++ b/llm-qa/llm_qa/vectorstores/init.py
--- a/llm-qa/llm_qa/vectorstores/qdrant.py
+++ b/llm-qa/llm_qa/vectorstores/qdrant.py
@ -1,22 +0,0 @@
-import logging
-
-from langchain.docstore.document import Document
-from langchain.embeddings.base import Embeddings
-from langchain.vectorstores.qdrant import Qdrant
-
-logger = logging.getLogger(__name__)
-
-
-async def upsert_documents(
-    documents: list[Document], embeddings: Embeddings, qdrant_url: str, collection: str
-) -> None:
-    logger.info(
-        "Upserting %d documents to Qdrant collection `%s`", len(documents), collection
-    )
-    await Qdrant.afrom_documents(
-        documents=documents,
-        embedding=embeddings,
-        url=qdrant_url,
-        prefer_grpc=True,
-        collection_name=collection,
-    )
--- a/llm-qa/llm_qa/web.py
+++ b/llm-qa/llm_qa/web.py
@ -1,5 +1,6 @@
 from fastapi import FastAPI

+from llm_qa.logging import load_logging_config
 from llm_qa.routers import api_v1

 app = FastAPI(title="LLM QA")
@ -9,4 +10,5 @@ app.include_router(api_v1.router)
 if __name__ == "__main__":
    import uvicorn

+    load_logging_config()
    uvicorn.run("llm_qa.web:app", host="0.0.0.0", port=8000, reload=True)  # noqa: S104
--- a/llm-qa/logging.yaml
+++ b/llm-qa/logging.yaml
@ -0,0 +1,27 @@
+version: 1
+disable_existing_loggers: False
+
+formatters:
+    simple:
+        format: "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
+
+handlers:
+    console:
+        class: logging.StreamHandler
+        level: DEBUG
+        formatter: simple
+        stream: ext://sys.stdout
+
+loggers:
+    uvicorn:
+        level: INFO
+        handlers: [console]
+        propagate: no
+    gunicorn:
+        level: INFO
+        handlers: [console]
+        propagate: no
+
+root:
+    level: INFO
+    handlers: [console]
--- a/llm-qa/poetry.lock
+++ b/llm-qa/poetry.lock
@ -528,6 +528,140 @@ files = [
 docs = ["Sphinx", "furo"]
 test = ["objgraph", "psutil"]

+[[package]]
+name = "grpcio"
+version = "1.60.1"
+description = "HTTP/2-based RPC framework"
+optional = false
+python-versions = ">=3.7"
+files = [
+    {file = "grpcio-1.60.1-cp310-cp310-linux_armv7l.whl", hash = "sha256:14e8f2c84c0832773fb3958240c69def72357bc11392571f87b2d7b91e0bb092"},
+    {file = "grpcio-1.60.1-cp310-cp310-macosx_12_0_universal2.whl", hash = "sha256:33aed0a431f5befeffd9d346b0fa44b2c01aa4aeae5ea5b2c03d3e25e0071216"},
+    {file = "grpcio-1.60.1-cp310-cp310-manylinux_2_17_aarch64.whl", hash = "sha256:fead980fbc68512dfd4e0c7b1f5754c2a8e5015a04dea454b9cada54a8423525"},
+    {file = "grpcio-1.60.1-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:082081e6a36b6eb5cf0fd9a897fe777dbb3802176ffd08e3ec6567edd85bc104"},
+    {file = "grpcio-1.60.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:55ccb7db5a665079d68b5c7c86359ebd5ebf31a19bc1a91c982fd622f1e31ff2"},
+    {file = "grpcio-1.60.1-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:9b54577032d4f235452f77a83169b6527bf4b77d73aeada97d45b2aaf1bf5ce0"},
+    {file = "grpcio-1.60.1-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:7d142bcd604166417929b071cd396aa13c565749a4c840d6c702727a59d835eb"},
+    {file = "grpcio-1.60.1-cp310-cp310-win32.whl", hash = "sha256:2a6087f234cb570008a6041c8ffd1b7d657b397fdd6d26e83d72283dae3527b1"},
+    {file = "grpcio-1.60.1-cp310-cp310-win_amd64.whl", hash = "sha256:f2212796593ad1d0235068c79836861f2201fc7137a99aa2fea7beeb3b101177"},
+    {file = "grpcio-1.60.1-cp311-cp311-linux_armv7l.whl", hash = "sha256:79ae0dc785504cb1e1788758c588c711f4e4a0195d70dff53db203c95a0bd303"},
+    {file = "grpcio-1.60.1-cp311-cp311-macosx_10_10_universal2.whl", hash = "sha256:4eec8b8c1c2c9b7125508ff7c89d5701bf933c99d3910e446ed531cd16ad5d87"},
+    {file = "grpcio-1.60.1-cp311-cp311-manylinux_2_17_aarch64.whl", hash = "sha256:8c9554ca8e26241dabe7951aa1fa03a1ba0856688ecd7e7bdbdd286ebc272e4c"},
+    {file = "grpcio-1.60.1-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:91422ba785a8e7a18725b1dc40fbd88f08a5bb4c7f1b3e8739cab24b04fa8a03"},
+    {file = "grpcio-1.60.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:cba6209c96828711cb7c8fcb45ecef8c8859238baf15119daa1bef0f6c84bfe7"},
+    {file = "grpcio-1.60.1-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:c71be3f86d67d8d1311c6076a4ba3b75ba5703c0b856b4e691c9097f9b1e8bd2"},
+    {file = "grpcio-1.60.1-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:af5ef6cfaf0d023c00002ba25d0751e5995fa0e4c9eec6cd263c30352662cbce"},
+    {file = "grpcio-1.60.1-cp311-cp311-win32.whl", hash = "sha256:a09506eb48fa5493c58f946c46754ef22f3ec0df64f2b5149373ff31fb67f3dd"},
+    {file = "grpcio-1.60.1-cp311-cp311-win_amd64.whl", hash = "sha256:49c9b6a510e3ed8df5f6f4f3c34d7fbf2d2cae048ee90a45cd7415abab72912c"},
+    {file = "grpcio-1.60.1-cp312-cp312-linux_armv7l.whl", hash = "sha256:b58b855d0071575ea9c7bc0d84a06d2edfbfccec52e9657864386381a7ce1ae9"},
+    {file = "grpcio-1.60.1-cp312-cp312-macosx_10_10_universal2.whl", hash = "sha256:a731ac5cffc34dac62053e0da90f0c0b8560396a19f69d9703e88240c8f05858"},
+    {file = "grpcio-1.60.1-cp312-cp312-manylinux_2_17_aarch64.whl", hash = "sha256:cf77f8cf2a651fbd869fbdcb4a1931464189cd210abc4cfad357f1cacc8642a6"},
+    {file = "grpcio-1.60.1-cp312-cp312-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:c557e94e91a983e5b1e9c60076a8fd79fea1e7e06848eb2e48d0ccfb30f6e073"},
+    {file = "grpcio-1.60.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:069fe2aeee02dfd2135d562d0663fe70fbb69d5eed6eb3389042a7e963b54de8"},
+    {file = "grpcio-1.60.1-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:cb0af13433dbbd1c806e671d81ec75bd324af6ef75171fd7815ca3074fe32bfe"},
+    {file = "grpcio-1.60.1-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:2f44c32aef186bbba254129cea1df08a20be414144ac3bdf0e84b24e3f3b2e05"},
+    {file = "grpcio-1.60.1-cp312-cp312-win32.whl", hash = "sha256:a212e5dea1a4182e40cd3e4067ee46be9d10418092ce3627475e995cca95de21"},
+    {file = "grpcio-1.60.1-cp312-cp312-win_amd64.whl", hash = "sha256:6e490fa5f7f5326222cb9f0b78f207a2b218a14edf39602e083d5f617354306f"},
+    {file = "grpcio-1.60.1-cp37-cp37m-linux_armv7l.whl", hash = "sha256:4216e67ad9a4769117433814956031cb300f85edc855252a645a9a724b3b6594"},
+    {file = "grpcio-1.60.1-cp37-cp37m-macosx_10_10_universal2.whl", hash = "sha256:73e14acd3d4247169955fae8fb103a2b900cfad21d0c35f0dcd0fdd54cd60367"},
+    {file = "grpcio-1.60.1-cp37-cp37m-manylinux_2_17_aarch64.whl", hash = "sha256:6ecf21d20d02d1733e9c820fb5c114c749d888704a7ec824b545c12e78734d1c"},
+    {file = "grpcio-1.60.1-cp37-cp37m-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:33bdea30dcfd4f87b045d404388469eb48a48c33a6195a043d116ed1b9a0196c"},
+    {file = "grpcio-1.60.1-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:53b69e79d00f78c81eecfb38f4516080dc7f36a198b6b37b928f1c13b3c063e9"},
+    {file = "grpcio-1.60.1-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:39aa848794b887120b1d35b1b994e445cc028ff602ef267f87c38122c1add50d"},
+    {file = "grpcio-1.60.1-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:72153a0d2e425f45b884540a61c6639436ddafa1829a42056aa5764b84108b8e"},
+    {file = "grpcio-1.60.1-cp37-cp37m-win_amd64.whl", hash = "sha256:50d56280b482875d1f9128ce596e59031a226a8b84bec88cb2bf76c289f5d0de"},
+    {file = "grpcio-1.60.1-cp38-cp38-linux_armv7l.whl", hash = "sha256:6d140bdeb26cad8b93c1455fa00573c05592793c32053d6e0016ce05ba267549"},
+    {file = "grpcio-1.60.1-cp38-cp38-macosx_10_10_universal2.whl", hash = "sha256:bc808924470643b82b14fe121923c30ec211d8c693e747eba8a7414bc4351a23"},
+    {file = "grpcio-1.60.1-cp38-cp38-manylinux_2_17_aarch64.whl", hash = "sha256:70c83bb530572917be20c21f3b6be92cd86b9aecb44b0c18b1d3b2cc3ae47df0"},
+    {file = "grpcio-1.60.1-cp38-cp38-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:9b106bc52e7f28170e624ba61cc7dc6829566e535a6ec68528f8e1afbed1c41f"},
+    {file = "grpcio-1.60.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:30e980cd6db1088c144b92fe376747328d5554bc7960ce583ec7b7d81cd47287"},
+    {file = "grpcio-1.60.1-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:0c5807e9152eff15f1d48f6b9ad3749196f79a4a050469d99eecb679be592acc"},
+    {file = "grpcio-1.60.1-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:f1c3dc536b3ee124e8b24feb7533e5c70b9f2ef833e3b2e5513b2897fd46763a"},
+    {file = "grpcio-1.60.1-cp38-cp38-win32.whl", hash = "sha256:d7404cebcdb11bb5bd40bf94131faf7e9a7c10a6c60358580fe83913f360f929"},
+    {file = "grpcio-1.60.1-cp38-cp38-win_amd64.whl", hash = "sha256:c8754c75f55781515a3005063d9a05878b2cfb3cb7e41d5401ad0cf19de14872"},
+    {file = "grpcio-1.60.1-cp39-cp39-linux_armv7l.whl", hash = "sha256:0250a7a70b14000fa311de04b169cc7480be6c1a769b190769d347939d3232a8"},
+    {file = "grpcio-1.60.1-cp39-cp39-macosx_10_10_universal2.whl", hash = "sha256:660fc6b9c2a9ea3bb2a7e64ba878c98339abaf1811edca904ac85e9e662f1d73"},
+    {file = "grpcio-1.60.1-cp39-cp39-manylinux_2_17_aarch64.whl", hash = "sha256:76eaaba891083fcbe167aa0f03363311a9f12da975b025d30e94b93ac7a765fc"},
+    {file = "grpcio-1.60.1-cp39-cp39-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:e5d97c65ea7e097056f3d1ead77040ebc236feaf7f71489383d20f3b4c28412a"},
+    {file = "grpcio-1.60.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2bb2a2911b028f01c8c64d126f6b632fcd8a9ac975aa1b3855766c94e4107180"},
+    {file = "grpcio-1.60.1-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:5a1ebbae7e2214f51b1f23b57bf98eeed2cf1ba84e4d523c48c36d5b2f8829ff"},
+    {file = "grpcio-1.60.1-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:9a66f4d2a005bc78e61d805ed95dedfcb35efa84b7bba0403c6d60d13a3de2d6"},
+    {file = "grpcio-1.60.1-cp39-cp39-win32.whl", hash = "sha256:8d488fbdbf04283f0d20742b64968d44825617aa6717b07c006168ed16488804"},
+    {file = "grpcio-1.60.1-cp39-cp39-win_amd64.whl", hash = "sha256:61b7199cd2a55e62e45bfb629a35b71fc2c0cb88f686a047f25b1112d3810904"},
+    {file = "grpcio-1.60.1.tar.gz", hash = "sha256:dd1d3a8d1d2e50ad9b59e10aa7f07c7d1be2b367f3f2d33c5fade96ed5460962"},
+]
+
+[package.extras]
+protobuf = ["grpcio-tools (>=1.60.1)"]
+
+[[package]]
+name = "grpcio-tools"
+version = "1.60.1"
+description = "Protobuf code generator for gRPC"
+optional = false
+python-versions = ">=3.7"
+files = [
+    {file = "grpcio-tools-1.60.1.tar.gz", hash = "sha256:da08224ab8675c6d464b988bd8ca02cccd2bf0275bceefe8f6219bfd4a4f5e85"},
+    {file = "grpcio_tools-1.60.1-cp310-cp310-linux_armv7l.whl", hash = "sha256:184b27333b627a7cc0972fb70d21a8bb7c02ac4a6febc16768d78ea8ff883ddd"},
+    {file = "grpcio_tools-1.60.1-cp310-cp310-macosx_12_0_universal2.whl", hash = "sha256:18d7737f29ef5bbe3352547d0eccd080807834f00df223867dfc860bf81e9180"},
+    {file = "grpcio_tools-1.60.1-cp310-cp310-manylinux_2_17_aarch64.whl", hash = "sha256:cc8ba358d2c658c6ecbc58e779bf0fc5a673fecac015a70db27fc5b4d37b76b6"},
+    {file = "grpcio_tools-1.60.1-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:2973f75e8ba5c551033a1d59cc97654f6f386deaf2559082011d245d7ed87bba"},
+    {file = "grpcio_tools-1.60.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:28ae665113affebdd109247386786e5ab4dccfcfad1b5f68e9cce2e326b57ee6"},
+    {file = "grpcio_tools-1.60.1-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:5c7ed086fef5ff59f46d53a052b1934b73e0f7d12365d656d6af3a88057d5a3e"},
+    {file = "grpcio_tools-1.60.1-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:8540f6480428a52614db71dd6394f52cbc0d2565b5ea1136a982f26390a42c7a"},
+    {file = "grpcio_tools-1.60.1-cp310-cp310-win32.whl", hash = "sha256:5b4a939097005531edec331f22d0b82bff26e71ede009354d2f375b5d41e74f0"},
+    {file = "grpcio_tools-1.60.1-cp310-cp310-win_amd64.whl", hash = "sha256:075bb67895970f96aabc1761ca674bf4db193f8fcad387f08e50402023b5f953"},
+    {file = "grpcio_tools-1.60.1-cp311-cp311-linux_armv7l.whl", hash = "sha256:284749d20fb22418f17d3d351b9eb838caf4a0393a9cb02c36e5c32fa4bbe9db"},
+    {file = "grpcio_tools-1.60.1-cp311-cp311-macosx_10_10_universal2.whl", hash = "sha256:b1041377cf32ee2338284ee26e6b9c10f9ea7728092376b19803dcb9b91d510d"},
+    {file = "grpcio_tools-1.60.1-cp311-cp311-manylinux_2_17_aarch64.whl", hash = "sha256:e529cd3d4109a6f4a3f7bdaca68946eb33734e2d7ffe861785a0586abe99ee67"},
+    {file = "grpcio_tools-1.60.1-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:31294b534f25f02ead204e58dcbe0e5437a95a1a6f276bb9378905595b02ff6d"},
+    {file = "grpcio_tools-1.60.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3fb6f4d2df0388c35c2804ba170f511238a681b679ead013bfe5e39d0ea9cf48"},
+    {file = "grpcio_tools-1.60.1-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:40cd8268a675269ce59c4fa50877597ec638bb1099c52237bb726c8ac9791868"},
+    {file = "grpcio_tools-1.60.1-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:985ac476da365267a2367ab20060f9096fbfc2e190fb02dd394f9ec05edf03ca"},
+    {file = "grpcio_tools-1.60.1-cp311-cp311-win32.whl", hash = "sha256:bd85f6c368b93ae45edf8568473053cb1cc075ef3489efb18f9832d4ecce062f"},
+    {file = "grpcio_tools-1.60.1-cp311-cp311-win_amd64.whl", hash = "sha256:c20e752ff5057758845f4e5c7a298739bfba291f373ed18ea9c7c7acbe69e8ab"},
+    {file = "grpcio_tools-1.60.1-cp312-cp312-linux_armv7l.whl", hash = "sha256:aafc94616c5f89c891d859057b194a153c451f9921053454e9d7d4cbf79047eb"},
+    {file = "grpcio_tools-1.60.1-cp312-cp312-macosx_10_10_universal2.whl", hash = "sha256:9bba347000f57dae8aea79c0d76ef7d72895597524d30d0170c7d1974a3a03f3"},
+    {file = "grpcio_tools-1.60.1-cp312-cp312-manylinux_2_17_aarch64.whl", hash = "sha256:1e96a532d38411f0543fe1903ff522f7142a9901afb0ed94de58d79caf1905be"},
+    {file = "grpcio_tools-1.60.1-cp312-cp312-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:5ea6e397d87f458bb2c387a4a6e1b65df74ce5b5194a1f16850c38309012e981"},
+    {file = "grpcio_tools-1.60.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3aeecd5b8faa2aab67e6c8b8a57e888c00ce70d39f331ede0a21312e92def1a6"},
+    {file = "grpcio_tools-1.60.1-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:d2c26ce5f774c98bd2d3d8d1703048394018b55d297ebdb41ed2ba35b9a34f68"},
+    {file = "grpcio_tools-1.60.1-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:214281cdafb7acfdcde848eca2de7c888a6e2b5cd25ab579712b965ea09a9cd4"},
+    {file = "grpcio_tools-1.60.1-cp312-cp312-win32.whl", hash = "sha256:8c4b917aa4fcdc77990773063f0f14540aab8d4a8bf6c862b964a45d891a31d2"},
+    {file = "grpcio_tools-1.60.1-cp312-cp312-win_amd64.whl", hash = "sha256:0aa34c7c21cff2177a4096b2b0d51dfbc9f8a41f929847a434e89b352c5a215d"},
+    {file = "grpcio_tools-1.60.1-cp37-cp37m-linux_armv7l.whl", hash = "sha256:acdba77584981fe799104aa545d9d97910bcf88c69b668b768c1f3e7d7e5afac"},
+    {file = "grpcio_tools-1.60.1-cp37-cp37m-macosx_10_10_universal2.whl", hash = "sha256:2a7fa55bc62d4b8ebe6fb26f8cf89df3cf3b504eb6c5f3a2f0174689d35fddb0"},
+    {file = "grpcio_tools-1.60.1-cp37-cp37m-manylinux_2_17_aarch64.whl", hash = "sha256:dffa326cf901fe08a0e218d9fdf593f12276088a8caa07fcbec7d051149cf9ef"},
+    {file = "grpcio_tools-1.60.1-cp37-cp37m-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:cf945bd22f396c0d0c691e0990db2bfc4e77816b1edc2aea8a69c35ae721aac9"},
+    {file = "grpcio_tools-1.60.1-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6801cfc5a85f0fb6fd12cade45942aaa1c814422328d594d12d364815fe34123"},
+    {file = "grpcio_tools-1.60.1-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:f95bdc6c7c50b7fc442e53537bc5b4eb8cab2a671c1da80d40b5a4ab1fd5d416"},
+    {file = "grpcio_tools-1.60.1-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:402efeec36d8b12b792bae8a900085416fc2f57a34b599445ace2e847b6b0d75"},
+    {file = "grpcio_tools-1.60.1-cp37-cp37m-win_amd64.whl", hash = "sha256:af88a2062b9c35034a80b25f289034b9c3c00c42bb88efaa465503a06fbd6a87"},
+    {file = "grpcio_tools-1.60.1-cp38-cp38-linux_armv7l.whl", hash = "sha256:46b495bae31c5d3f6ac0240eb848f0642b5410f80dff2aacdea20cdea3938c1d"},
+    {file = "grpcio_tools-1.60.1-cp38-cp38-macosx_10_10_universal2.whl", hash = "sha256:b5ae375207af9aa82f516dcd513d2e0c83690b7788d45844daad846ed87550f8"},
+    {file = "grpcio_tools-1.60.1-cp38-cp38-manylinux_2_17_aarch64.whl", hash = "sha256:15f13e8f3d77b96adcb1e3615acec5b100bd836c6010c58a51465bcb9c06d128"},
+    {file = "grpcio_tools-1.60.1-cp38-cp38-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:c354505e6a3d170da374f20404ea6a78135502df4f5534e5c532bdf24c4cc2a5"},
+    {file = "grpcio_tools-1.60.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a8cfab27ba2bd36a3e3b522aed686133531e8b919703d0247a0885dae8815317"},
+    {file = "grpcio_tools-1.60.1-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:b6ef213cb0aecb2832ee82a2eac32f29f31f50b17ce020604d82205096a6bd0c"},
+    {file = "grpcio_tools-1.60.1-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:0b62cb2d43a7f0eacc6a6962dfff7c2564874012e1a72ae4167e762f449e2912"},
+    {file = "grpcio_tools-1.60.1-cp38-cp38-win32.whl", hash = "sha256:3fcabf484720a9fa1690e2825fc940027a05a0c79a1075a730008ef634bd8ad2"},
+    {file = "grpcio_tools-1.60.1-cp38-cp38-win_amd64.whl", hash = "sha256:22ce3e3d861321d208d8bfd6161ab976623520b179712c90b2c175151463a6b1"},
+    {file = "grpcio_tools-1.60.1-cp39-cp39-linux_armv7l.whl", hash = "sha256:4e66fe204da15e08e599adb3060109a42927c0868fe8933e2d341ea649eceb03"},
+    {file = "grpcio_tools-1.60.1-cp39-cp39-macosx_10_10_universal2.whl", hash = "sha256:c1047bd831de5d9da761e9dc246988d5f07d722186938dfd5f34807398101010"},
+    {file = "grpcio_tools-1.60.1-cp39-cp39-manylinux_2_17_aarch64.whl", hash = "sha256:eba5fafd70585fbd4cb6ae45e3c5e11d8598e2426c9f289b78f682c0606e81cb"},
+    {file = "grpcio_tools-1.60.1-cp39-cp39-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:bba7230c60238c7a4ffa29f1aff6d78edb41f2c79cbe4443406472b1c80ccb5d"},
+    {file = "grpcio_tools-1.60.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a2bb8efc2cd64bd8f2779b426dd7e94e60924078ba5150cbbb60a846e62d1ed2"},
+    {file = "grpcio_tools-1.60.1-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:26f91161a91f1601777751230eaaafdf416fed08a15c3ba2ae391088e4a906c6"},
+    {file = "grpcio_tools-1.60.1-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:2c19be2bba5583e30f88bb5d71b430176c396f0d6d0db3785e5845bfa3d28cd2"},
+    {file = "grpcio_tools-1.60.1-cp39-cp39-win32.whl", hash = "sha256:9aadc9c00baa2064baa4414cff7c269455449f14805a355226674d89c507342c"},
+    {file = "grpcio_tools-1.60.1-cp39-cp39-win_amd64.whl", hash = "sha256:652b08c9fef39186ce4f97f05f5440c0ed41f117db0f7d6cb0e0d75dbc6afd3f"},
+]
+
+[package.dependencies]
+grpcio = ">=1.60.1"
+protobuf = ">=4.21.6,<5.0dev"
+setuptools = "*"
+
 [[package]]
 name = "h11"
 version = "0.14.0"
@ -539,6 +673,32 @@ files = [
    {file = "h11-0.14.0.tar.gz", hash = "sha256:8f19fbbe99e72420ff35c00b27a34cb9937e902a8b810e2c88300c6f0a3b699d"},
 ]

+[[package]]
+name = "h2"
+version = "4.1.0"
+description = "HTTP/2 State-Machine based protocol implementation"
+optional = false
+python-versions = ">=3.6.1"
+files = [
+    {file = "h2-4.1.0-py3-none-any.whl", hash = "sha256:03a46bcf682256c95b5fd9e9a99c1323584c3eec6440d379b9903d709476bc6d"},
+    {file = "h2-4.1.0.tar.gz", hash = "sha256:a83aca08fbe7aacb79fec788c9c0bac936343560ed9ec18b82a13a12c28d2abb"},
+]
+
+[package.dependencies]
+hpack = ">=4.0,<5"
+hyperframe = ">=6.0,<7"
+
+[[package]]
+name = "hpack"
+version = "4.0.0"
+description = "Pure-Python HPACK header compression"
+optional = false
+python-versions = ">=3.6.1"
+files = [
+    {file = "hpack-4.0.0-py3-none-any.whl", hash = "sha256:84a076fad3dc9a9f8063ccb8041ef100867b1878b25ef0ee63847a5d53818a6c"},
+    {file = "hpack-4.0.0.tar.gz", hash = "sha256:fc41de0c63e687ebffde81187a948221294896f6bdc0ae2312708df339430095"},
+]
+
 [[package]]
 name = "httpcore"
 version = "1.0.2"
@ -574,6 +734,7 @@ files = [
 [package.dependencies]
 anyio = "*"
 certifi = "*"
+h2 = {version = ">=3,<5", optional = true, markers = "extra == \"http2\""}
 httpcore = "==1.*"
 idna = "*"
 sniffio = "*"
@ -584,6 +745,17 @@ cli = ["click (==8.*)", "pygments (==2.*)", "rich (>=10,<14)"]
 http2 = ["h2 (>=3,<5)"]
 socks = ["socksio (==1.*)"]

+[[package]]
+name = "hyperframe"
+version = "6.0.1"
+description = "HTTP/2 framing layer for Python"
+optional = false
+python-versions = ">=3.6.1"
+files = [
+    {file = "hyperframe-6.0.1-py3-none-any.whl", hash = "sha256:0ec6bafd80d8ad2195c4f03aacba3a8265e57bc4cff261e802bf39970ed02a15"},
+    {file = "hyperframe-6.0.1.tar.gz", hash = "sha256:ae510046231dc8e9ecb1a6586f63d2347bf4c8905914aa84ba585ae85f28a914"},
+]
+
 [[package]]
 name = "idna"
 version = "3.6"
@ -1053,6 +1225,25 @@ files = [
 [package.dependencies]
 ptyprocess = ">=0.5"

+[[package]]
+name = "portalocker"
+version = "2.8.2"
+description = "Wraps the portalocker recipe for easy usage"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "portalocker-2.8.2-py3-none-any.whl", hash = "sha256:cfb86acc09b9aa7c3b43594e19be1345b9d16af3feb08bf92f23d4dce513a28e"},
+    {file = "portalocker-2.8.2.tar.gz", hash = "sha256:2b035aa7828e46c58e9b31390ee1f169b98e1066ab10b9a6a861fe7e25ee4f33"},
+]
+
+[package.dependencies]
+pywin32 = {version = ">=226", markers = "platform_system == \"Windows\""}
+
+[package.extras]
+docs = ["sphinx (>=1.7.1)"]
+redis = ["redis"]
+tests = ["pytest (>=5.4.1)", "pytest-cov (>=2.8.1)", "pytest-mypy (>=0.8.0)", "pytest-timeout (>=2.1.0)", "redis", "sphinx (>=6.0.0)", "types-redis"]
+
 [[package]]
 name = "prompt-toolkit"
 version = "3.0.43"
@ -1067,6 +1258,26 @@ files = [
 [package.dependencies]
 wcwidth = "*"

+[[package]]
+name = "protobuf"
+version = "4.25.2"
+description = ""
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "protobuf-4.25.2-cp310-abi3-win32.whl", hash = "sha256:b50c949608682b12efb0b2717f53256f03636af5f60ac0c1d900df6213910fd6"},
+    {file = "protobuf-4.25.2-cp310-abi3-win_amd64.whl", hash = "sha256:8f62574857ee1de9f770baf04dde4165e30b15ad97ba03ceac65f760ff018ac9"},
+    {file = "protobuf-4.25.2-cp37-abi3-macosx_10_9_universal2.whl", hash = "sha256:2db9f8fa64fbdcdc93767d3cf81e0f2aef176284071507e3ede160811502fd3d"},
+    {file = "protobuf-4.25.2-cp37-abi3-manylinux2014_aarch64.whl", hash = "sha256:10894a2885b7175d3984f2be8d9850712c57d5e7587a2410720af8be56cdaf62"},
+    {file = "protobuf-4.25.2-cp37-abi3-manylinux2014_x86_64.whl", hash = "sha256:fc381d1dd0516343f1440019cedf08a7405f791cd49eef4ae1ea06520bc1c020"},
+    {file = "protobuf-4.25.2-cp38-cp38-win32.whl", hash = "sha256:33a1aeef4b1927431d1be780e87b641e322b88d654203a9e9d93f218ee359e61"},
+    {file = "protobuf-4.25.2-cp38-cp38-win_amd64.whl", hash = "sha256:47f3de503fe7c1245f6f03bea7e8d3ec11c6c4a2ea9ef910e3221c8a15516d62"},
+    {file = "protobuf-4.25.2-cp39-cp39-win32.whl", hash = "sha256:5e5c933b4c30a988b52e0b7c02641760a5ba046edc5e43d3b94a74c9fc57c1b3"},
+    {file = "protobuf-4.25.2-cp39-cp39-win_amd64.whl", hash = "sha256:d66a769b8d687df9024f2985d5137a337f957a0916cf5464d1513eee96a63ff0"},
+    {file = "protobuf-4.25.2-py3-none-any.whl", hash = "sha256:a8b7a98d4ce823303145bf3c1a8bdb0f2f4642a414b196f04ad9853ed0c8f830"},
+    {file = "protobuf-4.25.2.tar.gz", hash = "sha256:fe599e175cb347efc8ee524bcd4b902d11f7262c0e569ececcb89995c15f0a5e"},
+]
+
 [[package]]
 name = "ptyprocess"
 version = "0.7.0"
@ -1246,6 +1457,29 @@ files = [
 [package.extras]
 cli = ["click (>=5.0)"]

+[[package]]
+name = "pywin32"
+version = "306"
+description = "Python for Window Extensions"
+optional = false
+python-versions = "*"
+files = [
+    {file = "pywin32-306-cp310-cp310-win32.whl", hash = "sha256:06d3420a5155ba65f0b72f2699b5bacf3109f36acbe8923765c22938a69dfc8d"},
+    {file = "pywin32-306-cp310-cp310-win_amd64.whl", hash = "sha256:84f4471dbca1887ea3803d8848a1616429ac94a4a8d05f4bc9c5dcfd42ca99c8"},
+    {file = "pywin32-306-cp311-cp311-win32.whl", hash = "sha256:e65028133d15b64d2ed8f06dd9fbc268352478d4f9289e69c190ecd6818b6407"},
+    {file = "pywin32-306-cp311-cp311-win_amd64.whl", hash = "sha256:a7639f51c184c0272e93f244eb24dafca9b1855707d94c192d4a0b4c01e1100e"},
+    {file = "pywin32-306-cp311-cp311-win_arm64.whl", hash = "sha256:70dba0c913d19f942a2db25217d9a1b726c278f483a919f1abfed79c9cf64d3a"},
+    {file = "pywin32-306-cp312-cp312-win32.whl", hash = "sha256:383229d515657f4e3ed1343da8be101000562bf514591ff383ae940cad65458b"},
+    {file = "pywin32-306-cp312-cp312-win_amd64.whl", hash = "sha256:37257794c1ad39ee9be652da0462dc2e394c8159dfd913a8a4e8eb6fd346da0e"},
+    {file = "pywin32-306-cp312-cp312-win_arm64.whl", hash = "sha256:5821ec52f6d321aa59e2db7e0a35b997de60c201943557d108af9d4ae1ec7040"},
+    {file = "pywin32-306-cp37-cp37m-win32.whl", hash = "sha256:1c73ea9a0d2283d889001998059f5eaaba3b6238f767c9cf2833b13e6a685f65"},
+    {file = "pywin32-306-cp37-cp37m-win_amd64.whl", hash = "sha256:72c5f621542d7bdd4fdb716227be0dd3f8565c11b280be6315b06ace35487d36"},
+    {file = "pywin32-306-cp38-cp38-win32.whl", hash = "sha256:e4c092e2589b5cf0d365849e73e02c391c1349958c5ac3e9d5ccb9a28e017b3a"},
+    {file = "pywin32-306-cp38-cp38-win_amd64.whl", hash = "sha256:e8ac1ae3601bee6ca9f7cb4b5363bf1c0badb935ef243c4733ff9a393b1690c0"},
+    {file = "pywin32-306-cp39-cp39-win32.whl", hash = "sha256:e25fd5b485b55ac9c057f67d94bc203f3f6595078d1fb3b458c9c28b7153a802"},
+    {file = "pywin32-306-cp39-cp39-win_amd64.whl", hash = "sha256:39b61c15272833b5c329a2989999dcae836b1eed650252ab1b7bfbe1d59f30f4"},
+]
+
 [[package]]
 name = "pyyaml"
 version = "6.0.1"
@ -1306,6 +1540,29 @@ files = [
    {file = "PyYAML-6.0.1.tar.gz", hash = "sha256:bfdf460b1736c775f2ba9f6a92bca30bc2095067b8a9d77876d1fad6cc3b4a43"},
 ]

+[[package]]
+name = "qdrant-client"
+version = "1.7.3"
+description = "Client library for the Qdrant vector search engine"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "qdrant_client-1.7.3-py3-none-any.whl", hash = "sha256:b062420ba55eb847652c7d2a26404fb1986bea13aa785763024013f96a7a915c"},
+    {file = "qdrant_client-1.7.3.tar.gz", hash = "sha256:7b809be892cdc5137ae80ea3335da40c06499ad0b0072b5abc6bad79da1d29fc"},
+]
+
+[package.dependencies]
+grpcio = ">=1.41.0"
+grpcio-tools = ">=1.41.0"
+httpx = {version = ">=0.14.0", extras = ["http2"]}
+numpy = {version = ">=1.26", markers = "python_version >= \"3.12\""}
+portalocker = ">=2.7.0,<3.0.0"
+pydantic = ">=1.10.8"
+urllib3 = ">=1.26.14,<3"
+
+[package.extras]
+fastembed = ["fastembed (==0.1.1)"]
+
 [[package]]
 name = "requests"
 version = "2.31.0"
@ -1353,6 +1610,22 @@ files = [
    {file = "ruff-0.2.1.tar.gz", hash = "sha256:3b42b5d8677cd0c72b99fcaf068ffc62abb5a19e71b4a3b9cfa50658a0af02f1"},
 ]

+[[package]]
+name = "setuptools"
+version = "69.1.0"
+description = "Easily download, build, install, upgrade, and uninstall Python packages"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "setuptools-69.1.0-py3-none-any.whl", hash = "sha256:c054629b81b946d63a9c6e732bc8b2513a7c3ea645f11d0139a2191d735c60c6"},
+    {file = "setuptools-69.1.0.tar.gz", hash = "sha256:850894c4195f09c4ed30dba56213bf7c3f21d86ed6bdaafb5df5972593bfc401"},
+]
+
+[package.extras]
+docs = ["furo", "jaraco.packaging (>=9.3)", "jaraco.tidelift (>=1.4)", "pygments-github-lexers (==0.0.5)", "rst.linker (>=1.9)", "sphinx (<7.2.5)", "sphinx (>=3.5)", "sphinx-favicon", "sphinx-inline-tabs", "sphinx-lint", "sphinx-notfound-page (>=1,<2)", "sphinx-reredirects", "sphinxcontrib-towncrier"]
+testing = ["build[virtualenv]", "filelock (>=3.4.0)", "flake8-2020", "ini2toml[lite] (>=0.9)", "jaraco.develop (>=7.21)", "jaraco.envs (>=2.2)", "jaraco.path (>=3.2.0)", "pip (>=19.1)", "pytest (>=6)", "pytest-checkdocs (>=2.4)", "pytest-cov", "pytest-enabler (>=2.2)", "pytest-home (>=0.5)", "pytest-mypy (>=0.9.1)", "pytest-perf", "pytest-ruff (>=0.2.1)", "pytest-timeout", "pytest-xdist", "tomli-w (>=1.0.0)", "virtualenv (>=13.0.0)", "wheel"]
+testing-integration = ["build[virtualenv] (>=1.0.3)", "filelock (>=3.4.0)", "jaraco.envs (>=2.2)", "jaraco.path (>=3.2.0)", "packaging (>=23.1)", "pytest", "pytest-enabler", "pytest-xdist", "tomli", "virtualenv (>=13.0.0)", "wheel"]
+
 [[package]]
 name = "six"
 version = "1.16.0"
@ -1527,6 +1800,38 @@ files = [
 docs = ["myst-parser", "pydata-sphinx-theme", "sphinx"]
 test = ["argcomplete (>=3.0.3)", "mypy (>=1.7.0)", "pre-commit", "pytest (>=7.0,<7.5)", "pytest-mock", "pytest-mypy-testing"]

+[[package]]
+name = "typer"
+version = "0.9.0"
+description = "Typer, build great CLIs. Easy to code. Based on Python type hints."
+optional = false
+python-versions = ">=3.6"
+files = [
+    {file = "typer-0.9.0-py3-none-any.whl", hash = "sha256:5d96d986a21493606a358cae4461bd8cdf83cbf33a5aa950ae629ca3b51467ee"},
+    {file = "typer-0.9.0.tar.gz", hash = "sha256:50922fd79aea2f4751a8e0408ff10d2662bd0c8bbfa84755a699f3bada2978b2"},
+]
+
+[package.dependencies]
+click = ">=7.1.1,<9.0.0"
+typing-extensions = ">=3.7.4.3"
+
+[package.extras]
+all = ["colorama (>=0.4.3,<0.5.0)", "rich (>=10.11.0,<14.0.0)", "shellingham (>=1.3.0,<2.0.0)"]
+dev = ["autoflake (>=1.3.1,<2.0.0)", "flake8 (>=3.8.3,<4.0.0)", "pre-commit (>=2.17.0,<3.0.0)"]
+doc = ["cairosvg (>=2.5.2,<3.0.0)", "mdx-include (>=1.4.1,<2.0.0)", "mkdocs (>=1.1.2,<2.0.0)", "mkdocs-material (>=8.1.4,<9.0.0)", "pillow (>=9.3.0,<10.0.0)"]
+test = ["black (>=22.3.0,<23.0.0)", "coverage (>=6.2,<7.0)", "isort (>=5.0.6,<6.0.0)", "mypy (==0.910)", "pytest (>=4.4.0,<8.0.0)", "pytest-cov (>=2.10.0,<5.0.0)", "pytest-sugar (>=0.9.4,<0.10.0)", "pytest-xdist (>=1.32.0,<4.0.0)", "rich (>=10.11.0,<14.0.0)", "shellingham (>=1.3.0,<2.0.0)"]
+
+[[package]]
+name = "types-pyyaml"
+version = "6.0.12.12"
+description = "Typing stubs for PyYAML"
+optional = false
+python-versions = "*"
+files = [
+    {file = "types-PyYAML-6.0.12.12.tar.gz", hash = "sha256:334373d392fde0fdf95af5c3f1661885fa10c52167b14593eb856289e1855062"},
+    {file = "types_PyYAML-6.0.12.12-py3-none-any.whl", hash = "sha256:c05bc6c158facb0676674b7f11fe3960db4f389718e19e62bd2b84d6205cfd24"},
+]
+
 [[package]]
 name = "typing-extensions"
 version = "4.9.0"
@ -1705,4 +2010,4 @@ multidict = ">=4.0"
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.12"
-content-hash = "7ffd6635973f31fbcbc5e962102b8cf12920f83e383dc0bb73732522db897a6f"
+content-hash = "b37d2542dbbbd85c8afc9a16dde50037737f87d250cec6c83e377e1266a19997"
--- a/llm-qa/pyproject.toml
+++ b/llm-qa/pyproject.toml
@ -13,11 +13,15 @@ fastapi = "^0.109.2"
 uvicorn = "^0.27.1"
 httpx = "^0.26.0"
 pydantic-settings = "^2.1.0"
+pydantic = "^2.6.1"
+qdrant-client = "^1.7.3"
+typer = "^0.9.0"

 [tool.poetry.group.dev.dependencies]
 ruff = "0.2.1"
 mypy = "^1.8.0"
 ipython = "^8.21.0"
+types-pyyaml = "^6.0.12.12"

 [build-system]
 requires = ["poetry-core"]
@ -70,4 +74,4 @@ lint.select = [
    "LOG",
    "RUF",
 ]
-lint.ignore = ["E501", "ANN101", "PLR0913", "PLR0917", "ISC001"]
+lint.ignore = ["E501", "ANN101", "PLR0913", "PLR0917", "ISC001", "TRY003"]