Similarity search with Redis returns unreleveant data

Hello Everyone,

I have an project with gpt-3.5. it is based on Redis vector store and RAG over Turkish PDF file

I use RecursiveTextSplitter with 1000 chunk and 100 overlap with Unstructered pdf loader.
Similarity and mmr search returns unreleavent data. How can i improve my ingestion technic better?

create_redisearch_retriever(llm_type,index_name, from_existing=False):
    embedding = EmbeddingFactory.create_embedding(llm_type)

    if from_existing:
        try:
            redis = Redis.from_existing_index(embedding=embedding, redis_url=os.environ["REDIS_URL"],
                                              index_name=index_name, schema="redis_schema.yaml")
        except ValueError:
            return None
    else:
        chunks = load_and_split_documents(os.path.join("uploads", "document.pdf"))
        redis = Redis(embedding=embedding, redis_url=os.environ["REDIS_URL"], index_name="guidelines")
        redis.drop_index(index_name="guidelines", delete_documents=True)
        await redis.aadd_documents(chunks)
        # redis.write_schema("redis_schema.yaml")

    return redis.as_retriever(search_type="mmr", search_kwargs={"k": 2, "distance_threshold": 0.9})