LlamaIndex for RAG: Integration & Indexing Services

LlamaIndex Integration for RAG Systems

Insurance company operators were spending up to 12 minutes searching for answers among 15,000 pages of policies and instructions. We implemented LlamaIndex — time dropped to 1.5 minutes, accuracy rose to 91%, and outdated document errors fell from 8% to 0.4%. Over 5 years, we've completed 20+ RAG projects on LlamaIndex in finance, insurance, and retail. We guarantee accuracy of at least 90% on your data. Average savings at scale reach up to 2 million rubles per year per department, with project payback in 3–4 months.

What Problems Does LlamaIndex Solve?

Scattered data sources: PDF, Word, HTML, databases. LlamaIndex connects 150+ formats through native loaders — no need to write adapters. Slow search: ordinary vector search doesn't understand compound queries. SubQuestionQueryEngine splits the question into parts and processes them in parallel. Missing context: LlamaIndex adds metadata (date, author, document type) and filters by it, excluding outdated or irrelevant sources.

How LlamaIndex Accelerates Search in Unstructured Data

LlamaIndex uses multi-level indexing. Documents are split into chunks (typically 512 tokens with 50% overlap). For each chunk, an embedding is generated (OpenAI text-embedding-3-small, 1536 dimensions). Vectors are stored in Qdrant or another store. On query, the LLM chooses a strategy: direct search, SubQuestionQueryEngine, or RouterQueryEngine — depending on complexity. A built-in reranker improves relevance of top-10 results.

Basic RAG with LlamaIndex

from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Settings
from llama_index.core.node_parser import SentenceSplitter
from llama_index.llms.openai import OpenAI
from llama_index.embeddings.openai import OpenAIEmbedding

# Global settings
Settings.llm = OpenAI(model="gpt-4o", temperature=0)
Settings.embed_model = OpenAIEmbedding(model="text-embedding-3-small")
Settings.node_parser = SentenceSplitter(chunk_size=512, chunk_overlap=50)

# Load documents
documents = SimpleDirectoryReader("./data", recursive=True).load_data()

# Create index
index = VectorStoreIndex.from_documents(documents)

# Query
query_engine = index.as_query_engine(similarity_top_k=5)
response = query_engine.query("What is the warranty period for equipment?")
print(response)
# Access sources
for node in response.source_nodes:
    print(f"Score: {node.score:.3f}, Source: {node.metadata.get('file_name')}")

Integration with Vector Stores

from llama_index.vector_stores.qdrant import QdrantVectorStore
from llama_index.core import StorageContext
import qdrant_client

# Connect to Qdrant
client = qdrant_client.QdrantClient(url="http://localhost:6333")
vector_store = QdrantVectorStore(client=client, collection_name="docs")
storage_context = StorageContext.from_defaults(vector_store=vector_store)

# Index into Qdrant
index = VectorStoreIndex.from_documents(
    documents,
    storage_context=storage_context,
    show_progress=True,
)

# Reload existing index
index = VectorStoreIndex.from_vector_store(vector_store)

Why Choose LlamaIndex for RAG?

LlamaIndex beats LangChain in tasks requiring deep document handling. Built-in SubQuestionQueryEngine and RouterQueryEngine don't need custom prompts — they're ready for complex queries immediately. IngestionPipeline caches processing, speeding up re-indexing by 60%. Additionally, LlamaIndex supports Retrieval-Augmented Fine-Tuning (LlamaIndex documentation): domain-specific embedding fine-tuning improves recall by 15–20%. In our projects, average response time for complex queries with SubQuestionQueryEngine is 40% faster than bare LangChain.

SubQuestionQueryEngine: Decomposing Complex Questions

from llama_index.core.query_engine import SubQuestionQueryEngine
from llama_index.core.tools import QueryEngineTool

# Create tools from different sources
financial_tool = QueryEngineTool.from_defaults(
    query_engine=financial_index.as_query_engine(),
    name="financial_data",
    description="Company financial metrics for the last three years",
)

contracts_tool = QueryEngineTool.from_defaults(
    query_engine=contracts_index.as_query_engine(),
    name="contracts",
    description="Supplier and customer contracts",
)

# SubQuestion engine automatically splits query into sub-queries
engine = SubQuestionQueryEngine.from_defaults(
    query_engine_tools=[financial_tool, contracts_tool],
    use_async=True,
)

response = engine.query(
    "Compare last quarter's revenue with budget and check for overdue payments on contracts"
)
# The agent creates 2 sub-queries and merges results

RouterQueryEngine: Routing Across Indexes

from llama_index.core.query_engine.router_query_engine import RouterQueryEngine
from llama_index.core.selectors import LLMSingleSelector

router_engine = RouterQueryEngine(
    selector=LLMSingleSelector.from_defaults(),
    query_engine_tools=[
        QueryEngineTool.from_defaults(
            query_engine=summary_index.as_query_engine(response_mode="tree_summarize"),
            description="For summarization questions about the overall document",
        ),
        QueryEngineTool.from_defaults(
            query_engine=vector_index.as_query_engine(),
            description="For searching specific facts and details",
        ),
    ],
)

IngestionPipeline: Advanced Preprocessing

from llama_index.core.ingestion import IngestionPipeline, IngestionCache
from llama_index.core.node_parser import SentenceSplitter, SemanticSplitterNodeParser
from llama_index.core.extractors import TitleExtractor, QuestionsAnsweredExtractor
from llama_index.core.vector_stores import SimpleVectorStore

pipeline = IngestionPipeline(
    transformations=[
        SentenceSplitter(chunk_size=512, chunk_overlap=64),
        TitleExtractor(nodes=3),  # Adds document title to each chunk's metadata
        QuestionsAnsweredExtractor(questions=5),  # Generates hypothetical questions for HyDE
        OpenAIEmbedding(model="text-embedding-3-small"),
    ],
    vector_store=vector_store,
    cache=IngestionCache(),  # Caches processed documents
)

nodes = await pipeline.arun(documents=documents, show_progress=True)

Practical Case: Enterprise Knowledge Base for an Insurance Company

Initial situation: 15,000 pages of documents (policies, insurance rules, regulatory instructions, internal regulations). Operators spent 8–12 minutes finding an answer to a customer query.

LlamaIndex architecture (our project):

• Sources: 4 document types in separate indexes in Qdrant
• RouterQueryEngine: routing by question type
• SubQuestionQueryEngine: for questions covering multiple types
• IngestionPipeline: automatic re-indexing on document updates
• Metadata filtering: by insurance type, document date, regional regulator

Results:

• Average operator response time: 10 min → 1.5 min
• Answer accuracy (expert evaluation): 91%
• Erroneous links to outdated policy versions: ~8% → 0.4%
• Document coverage: 73% (previously, operators were unaware of many documents)

LlamaIndex vs LangChain for RAG

Typical Implementation Scenarios for LlamaIndex

What's Included in the Work

1. Data source audit — identify document types, volume, update frequency.
2. Index design — choose chunker, embedding model, vector store.
3. Retrieval pipeline setup — configure RouterQueryEngine, SubQuestionQueryEngine, reranking.
4. Infrastructure integration — connect API, CI/CD, monitoring dashboard (latency p99, recall).
5. Team training — documentation and workshop on using the system.

Estimated Timelines

• Basic RAG with LlamaIndex: from 3 to 5 days
• Multi-source RAG with RouterQueryEngine: from 1 to 2 weeks
• IngestionPipeline with automatic updates: from 1 week
• Domain-specific embedding fine-tuning: from 2 to 3 weeks

Exact timelines are calculated after a data audit. We'll assess your project in 1 day — contact us. Request a data audit and receive a commercial proposal. We guarantee payback in 3–4 months through reduced search time: savings up to 2 million rubles per year on operators.

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.