public class RetrievalMarkdownExample {

    public static final String NAMESPACE = "markdown";

    public static void main(String[] args) {
        // Load Notion page as a markdown file
        String path = "docs/extras/use_cases/question_answering/notion_db/";
        var loader = new NotionDirectoryLoader(path);
        var docs = loader.load();
        var mdFile = docs.get(0).getPageContent();

        // Let's create groups based on the section headers in our page
        List<Pair<String, String>> headersToSplitOn = List.of(Pair.of("###", "Section"));
        MarkdownHeaderTextSplitter markdownSplitter = new MarkdownHeaderTextSplitter(headersToSplitOn);
        List<Document> mdHeaderSplits = markdownSplitter.splitText(mdFile);

        // Define our text splitter
        var textSplitter = RecursiveCharacterTextSplitter.builder()
                .chunkSize(500)
                .chunkOverlap(0)
                .keepSeparator(true)
                .build();
        var allSplits = textSplitter.splitDocuments(mdHeaderSplits);

        // Build pinecone and keep the metadata
        var vectorStore = initializePineconeIndex(NAMESPACE, allSplits);

        // Define our metadata
        var metadataFieldInfo = List.of(
                new AttributeInfo("Section", "Part of the document that the text comes from",
                        "string or list[string]"));
        var documentContentDescription = "Major sections of the document";

        // Define self query retriever
        var llm = OpenAI.builder().temperature(0).requestTimeout(30).build().init();
        var retriever = SelfQueryRetriever.fromLLM(llm, vectorStore, documentContentDescription, metadataFieldInfo);

        // create chat or Q+A apps that are aware of the explicit document structure.
        var chat = ChatOpenAI.builder().temperature(0).build().init();
        var qaChain = RetrievalQa.fromChainType(chat, retriever);
        var result = qaChain.run("Summarize the Testing section of the document");
        println(result);
    }
}