Automated RAG Quality Evaluation with RAGAS

We are a team of AI engineers with over 7 years of experience in NLP and 30+ RAG systems delivered. Guaranteed metrics improvement or a detailed report. Without systematic RAGAS evaluation, the RAG system remains a black box: you don't know how many hallucinations the model generates or how much relevant context is lost. RAGAS (RAG Assessment) is the most popular framework for automated evaluation, using an LLM as a judge and requiring no manual labeling. On one project, we reduced manual answer verification by 80%, saving the team over $30,000 per year in evaluation costs and significantly cutting the evaluation budget. In 3–6 weeks, we set up the full evaluation pipeline, integrate it into CI/CD, and drive metrics to 0.9+. Contact us for an audit of your RAG system.

What is RAGAS and why do you need it?

RAGAS is an open-source framework for evaluating the quality of RAG systems. It automatically generates test sets from your documents and evaluates LLM answers across five key metrics. Without RAGAS, you evaluate quality manually or not at all, leading to uncontrolled hallucinations and context loss. RAGAS makes evaluation objective and reproducible. Learn more on GitHub.

RAGAS Metrics

How to interpret RAGAS metrics?

Context Precision below 0.7 signals that the system retrieves a lot of irrelevant context. Solutions: improve reranking, add metadata filtering, reduce top_k.

Context Recall below 0.7: the system fails to find the needed documents. Solutions: improve chunking, try hybrid search, fine-tune embedding models.

Faithfulness below 0.8 is a critical signal: the model hallucinates, making up information not supported by the context. Solutions: improve system prompt, add an instruction to answer only based on context, lower temperature. This is the most important metric for legal and medical RAG systems.

Answer Relevancy below 0.8: answers are off-topic. Solutions: improve prompt, add few-shot examples of the desired format.

How to implement RAGAS in 4 steps

1. Prepare your documents and create a test set (at least 200 questions). RAGAS will automatically generate questions of varying difficulty.
2. Run a baseline evaluation on all metrics. This takes 1–2 days.
3. Analyze the results and perform optimization iterations (improve retrieval, prompt, chunking).
4. Integrate continuous evaluation into CI/CD — now every commit will be checked for quality degradation.

Setting up the pipeline takes 2–3 days, test generation 1–2 days, and the full cycle with improvements 3 to 6 weeks.

Installation and basic usage of RAGAS

from ragas import evaluate
from ragas.metrics import (
    context_precision,
    context_recall,
    faithfulness,
    answer_relevancy,
    answer_correctness,
)
from datasets import Dataset

# Prepare dataset for evaluation
eval_data = {
    "question": [
        "What is the contract duration?",
        "Who is responsible for delivery delay?",
    ],
    "answer": [
        "The contract is valid until December 31.",
        "The supplier is responsible for delays exceeding 5 business days.",
    ],
    "contexts": [
        ["2.1. This Agreement comes into force upon signing and is valid until December 31..."],
        ["4.3. In case of delivery delay beyond 5 business days, the Supplier..."],
    ],
    "ground_truths": [
        "The contract is valid until December 31.",
        "The Supplier is responsible for delays exceeding 5 business days.",
    ],
}

dataset = Dataset.from_dict(eval_data)

# Evaluation
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from ragas.llms import LangchainLLMWrapper
from ragas.embeddings import LangchainEmbeddingsWrapper

evaluator_llm = LangchainLLMWrapper(ChatOpenAI(model="gpt-4o"))
evaluator_embeddings = LangchainEmbeddingsWrapper(OpenAIEmbeddings())

results = evaluate(
    dataset,
    metrics=[context_precision, context_recall, faithfulness, answer_relevancy],
    llm=evaluator_llm,
    embeddings=evaluator_embeddings,
)

print(results)
# {'context_precision': 0.88, 'context_recall': 0.82, 'faithfulness': 0.94, 'answer_relevancy': 0.91}

Automated test set: Testset Generation

RAGAS can generate a test set from your documents:

from ragas.testset.generator import TestsetGenerator
from ragas.testset.evolutions import simple, reasoning, multi_context

generator = TestsetGenerator.with_openai()

# Generate tests of varying difficulty
testset = generator.generate_with_langchain_docs(
    documents=your_documents,
    test_size=100,
    distributions={
        simple: 0.5,          # Simple questions from one document
        reasoning: 0.3,       # Requires reasoning
        multi_context: 0.2,   # Requires multiple documents
    }
)

testset.to_pandas().to_csv("evaluation_testset.csv", index=False)

Practical case: iterations on a legal chatbot RAG system

From our practice: a client — a legal firm with a corpus of 10,000 contracts. The initial RAG on GPT-4o-mini and ChromaDB showed low metrics. We performed three optimization iterations.

Initial state (v1):

Iteration 1: added hybrid search (sparse + dense). Context Recall increased from 0.68 to 0.81 (+19%).

Iteration 2: added Contextual Compression and reranker. Context Precision improved from 0.61 to 0.84 (+38%), Faithfulness from 0.74 to 0.91 (+23%).

Iteration 3: refined system prompt with explicit anti-hallucination instructions. Faithfulness reached 0.95, Answer Relevancy 0.88.

Final state (v4):

RAGAS allowed us to automate evaluation, making it 10x faster than manual reviewer assessment. The budget savings on evaluation amounted to tens of thousands of dollars on each project.

Continuous Evaluation in CI/CD

import pytest

@pytest.fixture(scope="session")
def rag_evaluation_results():
    """Run RAGAS evaluation on the test set"""
    return evaluate(evaluation_dataset, metrics=[faithfulness, context_recall])

def test_faithfulness_above_threshold(rag_evaluation_results):
    assert rag_evaluation_results["faithfulness"] >= 0.85, \
        f"Faithfulness {rag_evaluation_results['faithfulness']:.2f} below threshold 0.85"

def test_context_recall_above_threshold(rag_evaluation_results):
    assert rag_evaluation_results["context_recall"] >= 0.75

What's included in the service

• Setting up the RAGAS pipeline tailored to your infrastructure.
• Generating a representative test set (200+ questions).
• Baseline evaluation on 5 metrics.
• 2–3 optimization iterations with specific recommendations.
• Integrating continuous evaluation into your CI/CD.
• Access to an evaluation dashboard and detailed documentation.
• A training session for your team (up to 2 hours).
• 3 months of email support for any follow-up questions.
• A report with results and a plan for further optimization.

Trusted by leading legal and financial institutions. Over 5 years of experience in RAG systems. We guarantee to improve your RAG metrics or provide a comprehensive diagnostic report at no extra cost.

Get a consultation on your RAG system evaluation — we will assess your project and propose a plan. Order a metrics audit now.

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.