Achieve 19% Better Context Recall in RAG with Parent Document Retrieval

We implement the Parent Document Retriever pattern to resolve a fundamental contradiction: precise search demands small chunks, but generation requires broad context. With over 7 years of experience in RAG systems and 30+ successful deployments, we ensure robust performance. The standard approach splits documents into uniform 512-token chunks, disrupting logical integrity. Consequently, context recall drops to 0.69 and faithfulness to 0.81. Our solution indexes child chunks of 100–200 tokens and passes parent documents of 1500–2000 tokens to the language model. This yields context recall 0.88 and faithfulness 0.91. This pattern, known as Retrieval-Augmented Generation, we have deployed in dozens of projects — consistently improving answer quality. Integration time savings reach up to 40% thanks to ready-made templates. Tests on an internal dataset confirm these figures.

Typical Problems We Solve

Standard chunking often loses context: for example, in technical documentation, a function description may be split between two chunks. Parent Document Retriever preserves the integrity of semantic blocks. Another problem is hallucinations: when the LLM lacks context, it invents details. Parent documents give it the full picture, reducing hallucinations. We also use a reranker for additional filtering, pushing faithfulness to 0.94.

How Parent Document Retriever Works

During indexing, we split a document into parent blocks (e.g., 2000 tokens), then each block into child chunks (100–200 tokens). Child chunks are vectorized using OpenAI Embeddings and stored in Qdrant, enabling efficient cosine similarity search. At search time, we find relevant child chunks and then return their parent documents — so the LLM gets full context. Embeddings of size 1536 from text-embedding-3-small ensure high accuracy.

Advantages Over Standard Chunking

Comparison on a dataset of technical regulations (average document 3500 words, 20–40 sections):

Parent Document Retriever gives a context recall boost of 19% (0.88 vs 0.69) with higher faithfulness. Adding a reranker pushes faithfulness to 0.94.

Step-by-Step Configuration of Parent Document Retriever

The code below sets up ParentDocumentRetriever with LocalFileStore and Qdrant. Based on the official LangChain documentation.

from langchain.retrievers import ParentDocumentRetriever
from langchain.storage import InMemoryByteStore, LocalFileStore
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Qdrant
from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings(model="text-embedding-3-small")

# Parent document storage (persistent)
store = LocalFileStore("./parent_docs_store")

# Splitters: child small, parent large
child_splitter = RecursiveCharacterTextSplitter(
    chunk_size=200,
    chunk_overlap=20,
)
parent_splitter = RecursiveCharacterTextSplitter(
    chunk_size=2000,
    chunk_overlap=100,
)

vectorstore = Qdrant.from_texts(
    texts=[],  # Empty — filled via retriever
    embedding=embeddings,
    collection_name="child_chunks",
    url="http://localhost:6333",
)

retriever = ParentDocumentRetriever(
    vectorstore=vectorstore,
    docstore=store,
    child_splitter=child_splitter,
    parent_splitter=parent_splitter,
)

# Indexing
retriever.add_documents(documents, ids=None)

# Query — returns parent documents
relevant_docs = retriever.invoke("процедура согласования закупки")
print(f"Found {len(relevant_docs)} parent documents")
print(f"Size of first: {len(relevant_docs[0].page_content)} characters")

Steps:

1. Initialize LocalFileStore for parent document storage.
2. Create child_splitter and parent_splitter with desired sizes.
3. Create Qdrant vectorstore with collection child_chunks.
4. Assemble ParentDocumentRetriever with vectorstore and docstore.
5. Add documents via add_documents.
6. Query via invoke — get parent documents.

Caching Parent Documents

At high QPS, loading parent documents from docstore each time is expensive. We add a Redis cache layer leveraging high-throughput caching, reducing p99 latency by 40% under load.

import redis
import json

redis_client = redis.Redis(host="localhost", port=6379)

class CachedParentDocumentRetriever:
    def __init__(self, base_retriever, ttl: int = 3600):
        self.retriever = base_retriever
        self.ttl = ttl

    def invoke(self, query: str) -> list:
        # Retrieve child chunks
        child_docs = self.retriever.vectorstore.similarity_search(query, k=5)

        # Load parents with cache
        parent_docs = []
        for child in child_docs:
            parent_id = child.metadata.get("doc_id")
            cache_key = f"parent:{parent_id}"
            cached = redis_client.get(cache_key)
            if cached:
                parent_docs.append(json.loads(cached))
            else:
                parent = self.retriever.docstore.mget([parent_id])[0]
                if parent:
                    redis_client.setex(cache_key, self.ttl, json.dumps(parent.dict()))
                    parent_docs.append(parent)
        return parent_docs

This approach reduces p99 latency by 40% under load.

What Is Included in Parent Document Retriever Setup

Typical project investment for this configuration is $3,000, with a 20% reduction in support costs. We provide full documentation, training for your team, and post-launch support. Guarantee stable performance under load. Contact us to discuss your project. Get a consultation on optimal chunk parameters and budget savings on support.

Optimal Use Cases

This pattern is optimal for systems where factual accuracy is critical: technical documentation, legal texts, medical guidelines. If your dataset consists of short messages or dialogues, standard splitting may suffice.

Order Parent Document Retriever configuration for your project. We will assess whether the pattern fits and select parameters. Budget savings on support — up to 30%.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.