63 lines
2.0 KiB
Python
63 lines
2.0 KiB
Python
"""Vector store utilities for the FAQ bot.
|
|
|
|
This module handles loading markdown files into a persistent Chroma vector store
|
|
and provides a tool that performs semantic search over the stored documents.
|
|
"""
|
|
|
|
import os
|
|
from pathlib import Path
|
|
from typing import List
|
|
|
|
from langchain_ollama import OllamaEmbeddings
|
|
from langchain_chroma import Chroma
|
|
from langchain_core.documents import Document
|
|
|
|
from config import DATA_DIR, CHROMA_PERSIST_DIR, CHROMA_TOP_K
|
|
from chunker import chunk_text
|
|
|
|
# Initialize embeddings once
|
|
embeddings = OllamaEmbeddings(model="nomic-embed-text")
|
|
|
|
# Create or load the Chroma collection
|
|
vector_store = Chroma(
|
|
collection_name="faq_collection",
|
|
embedding_function=embeddings,
|
|
persist_directory=CHROMA_PERSIST_DIR,
|
|
)
|
|
|
|
def load_faq_to_chroma() -> None:
|
|
"""Load all markdown files from DATA_DIR into the Chroma vector store.
|
|
|
|
The function reads each .md file, splits it into chunks, creates Document
|
|
objects, and adds them to the vector store. The store is persisted to disk.
|
|
"""
|
|
docs: List[Document] = []
|
|
for md_file in Path(DATA_DIR).glob("*.md"):
|
|
text = md_file.read_text(encoding="utf-8")
|
|
chunks = chunk_text(text)
|
|
for i, chunk in enumerate(chunks):
|
|
doc = Document(page_content=chunk, metadata={"source": md_file.name, "chunk_id": i})
|
|
docs.append(doc)
|
|
if docs:
|
|
vector_store.add_documents(docs)
|
|
vector_store.persist()
|
|
|
|
def search_course_docs(query: str, k: int = CHROMA_TOP_K) -> str:
|
|
"""Perform a semantic search over the FAQ collection.
|
|
|
|
Parameters
|
|
----------
|
|
query: str
|
|
The user question.
|
|
k: int, optional
|
|
Number of top results to return.
|
|
|
|
Returns
|
|
-------
|
|
str
|
|
Concatenated top-k documents' page content.
|
|
"""
|
|
results = vector_store.similarity_search_with_score(query, k=k)
|
|
# results is list of (Document, score)
|
|
snippets = [doc.page_content for doc, _ in results]
|
|
return "\n\n---\n\n".join(snippets) |