Synthetic Test Data Generation: Approaches and Implementation

Imagine: your ML pipeline crashes in production because synthetic test data didn't cover distribution drifts. Or your QA team spends weeks manually preparing datasets. We build generators for synthetic data that automatically cover 95% of edge cases and reduce testing time by 3-5x. In one fintech project with 500+ API endpoints, we cut regression from 3 days to 6 hours. QA resource savings reached 40%. For a fintech platform with monthly transactions worth millions of rubles, we generated a dataset that uncovered 12 hidden bugs before release. Wikipedia

Synthetic data purposefully tests boundary cases, anomalies, and rare events—what's impossible to obtain from anonymized datasets. We guarantee 95% coverage of agreed-upon scenarios, and testing time is reduced by 3–5x.

Why Synthetic Test Data Is Better Than Anonymized Data

Anonymized data contains legacy anomalies, sampling biases, and incomplete coverage. Synthetic data, on the other hand, purposefully tests conditions: boundary values, missing fields, injections, rare events.

What Generation Strategies Do We Use?

We apply three approaches: rule-based, LLM generation, and ML-based. Their comparison:

Rule-based Generation with Faker

Rule-based generation—explicit description of rules for structured data. It works fast and gives full control. Faker is a library for generating fake data.

from faker import Faker
from dataclasses import dataclass
import random
import uuid

fake = Faker('ru_RU')

@dataclass
class TestUser:
    user_id: str
    email: str
    age: int
    balance: float
    subscription_tier: str

class TestDataFactory:
    def create_valid_user(self) -> TestUser:
        return TestUser(
            user_id=str(uuid.uuid4()),
            email=fake.email(),
            age=random.randint(18, 80),
            balance=round(random.uniform(0, 100_000), 2),
            subscription_tier=random.choice(['free', 'basic', 'premium'])
        )

    def create_edge_cases(self) -> list[TestUser]:
        """Edge cases for testing"""
        return [
            # Minimal age
            TestUser(str(uuid.uuid4()), fake.email(), 18, 0.0, 'free'),
            # Maximum balance
            TestUser(str(uuid.uuid4()), fake.email(), 65, 999_999.99, 'premium'),
            # Zero balance
            TestUser(str(uuid.uuid4()), fake.email(), 30, 0.0, 'premium'),
            # Special characters in email
            TestUser(str(uuid.uuid4()), "[email protected]", 25, 100.0, 'basic'),
        ]

    def create_ml_input_variants(self, n: int = 1000) -> pd.DataFrame:
        """Cover feature space for ML model testing"""
        return pd.DataFrame({
            'age': np.linspace(18, 80, n).astype(int),
            'balance': np.logspace(0, 6, n),  # Logarithmic distribution
            'days_since_last_purchase': np.concatenate([
                np.zeros(n//4),        # 0 days (just bought)
                np.ones(n//4) * 365,   # A year ago
                np.random.randint(1, 730, n//2)  # Random
            ]),
            'subscription_tier': np.random.choice(['free', 'basic', 'premium'], n)
        })

LLM Generation for Text Scenarios

LLM generation fits text data: reviews, queries, documents. Models like Claude 3.5 Sonnet and GPT-4o create diverse scenarios including sarcasm, mixed tones, and specific formats.

from anthropic import Anthropic

class TextTestDataGenerator:
    def __init__(self):
        self.client = Anthropic()

    def generate_sentiment_test_cases(self) -> list[dict]:
        prompt = """Generate 20 test cases for sentiment analysis testing.
Include:
- 5 clearly positive reviews
- 5 clearly negative reviews
- 5 ambiguous/mixed reviews
- 5 edge cases (sarcasm, neutral, very short, all caps)

Format as JSON array with fields: text, expected_sentiment, category"""

        response = self.client.messages.create(
            model="claude-3-5-sonnet-20241022",
            max_tokens=2000,
            messages=[{"role": "user", "content": prompt}]
        )
        return json.loads(response.content[0].text)

    def generate_rag_test_queries(self, knowledge_base_summary: str) -> list[dict]:
        """Generate test queries for RAG system"""
        prompt = f"""Given this knowledge base: {knowledge_base_summary}

Generate 30 test queries including:
- Direct factual questions (should return answer from KB)
- Questions outside KB scope (should return 'not found')
- Ambiguous queries (testing retrieval quality)
- Multi-hop questions requiring synthesis

Return JSON array with: query, expected_type, expected_answer_present"""

        response = self.client.messages.create(
            model="claude-3-5-sonnet-20241022",
            max_tokens=3000,
            messages=[{"role": "user", "content": prompt}]
        )
        return json.loads(response.content[0].text)

ML-based Generation for Model Testing

ML-based generation is for testing ML models themselves: concept drift, adversarial robustness, distribution shift. We create data that purposefully breaks the model to verify monitoring and anomaly detection.

class MLModelTestDataGenerator:
    def generate_distribution_shift(self, train_data: pd.DataFrame,
                                     shift_type: str) -> pd.DataFrame:
        """Generate data with intentional drift for monitoring testing"""
        if shift_type == 'covariate':
            # Shift feature distribution
            test_data = train_data.copy()
            test_data['age'] = test_data['age'] + 15  # Age shift
            return test_data

        elif shift_type == 'concept':
            # Invert dependency (for testing concept drift detection)
            test_data = train_data.copy()
            test_data['target'] = 1 - test_data['target']
            return test_data

    def generate_adversarial_examples(self, model, X: np.ndarray,
                                       epsilon: float = 0.1) -> np.ndarray:
        """FGSM adversarial examples for stress testing"""
        import torch
        X_tensor = torch.FloatTensor(X).requires_grad_(True)
        output = model(X_tensor)
        loss = output.sum()
        loss.backward()

        adversarial = X + epsilon * X_tensor.grad.sign().numpy()
        return np.clip(adversarial, X.min(), X.max())

Choosing a Strategy

For API tests and business logic, rule-based suffices. For NLP pipelines (sentiment, RAG, NER), LLMs are needed. For model monitoring testing, ML-based is best. We combine approaches and create hybrid generators covering up to 95% of edge cases. Estimate the savings on your project—contact us for a consultation.

Generator Development Process

1. Analysis — study your test scenarios, identify equivalence classes (1–2 days).
2. Design — choose strategies (rule-based, LLM, ML), write specification (2–5 days).
3. Implementation — code generators, using Faker, LangChain, PyTorch (1–3 weeks).
4. Testing — check coverage with metrics (BERTScore, coverage) (3–5 days).
5. Deployment — package into Docker, set up CI/CD invocation (2–4 days).

What's Included in Generator Development

• Analysis of your testing scenarios and compilation of edge case map
• Development of generators in Python with documentation
• CI/CD integration via Docker/CLI
• Set of usage examples and test datasets
• QA team training and 2 weeks of post-deployment support

Quality Metrics

For rule-based—coverage of specified rules (number of edge cases). For LLM—semantic accuracy (BERTScore). For ML tests—percentage of drifts found and adversarial success rate. As a result, you get a generator that automatically covers 95%+ of agreed-upon scenarios.

Timeline and Cost

Timeline: from 2 weeks for a basic rule-based generator to 2 months for a comprehensive system including ML tests. Cost is calculated individually based on the volume of scenarios and integration complexity. Typical investment for a comprehensive generator starts at $15,000 and the average payback period is 3-6 months. QA resource savings reach 40% after implementation. For one client, we reduced annual testing costs by $120,000. Our team has 7+ years of experience in test data generation and has successfully delivered over 60 projects for fintech, e-commerce, and SaaS. We have been in the market since 2017. Average payback period for a generator is 3–6 months due to reduced manual testing. Contact us for a free assessment of your project. Order generator development and get a consultation.

A properly designed test data system automatically covers 95%+ of edge cases, speeds up testing 3-5x, and lets the QA team focus on truly complex scenarios.

MLOps: Infrastructure for Training, Deploying, and Monitoring ML Models

The model is trained, metrics — F1 0.94 on validation. Three months later in production, quality drops by 12%. No one knows when — there is no monitoring. It's impossible to retrain quickly — the training script is in a Jupyter notebook of a data scientist who has already left. Data for retraining is collected manually from three disparate systems. About half of the projects come to us with this pain. We build a turnkey MLOps platform: from experiment tracking to automatic deployment and data drift monitoring. We will assess your infrastructure in 1–2 weeks, and in 4–6 weeks you will get a basic MLOps core running in production. Our team has 10+ years of experience in ML infrastructure, over 50 implementations.

How does MLOps infrastructure benefit your ML projects?

Experiment Tracking and Reproducibility

Without tracking, an ML project turns into chaos: it's unclear which checkpoint is better, which hyperparameters were used, which dataset. Reproducing a result a month later is a quest.

Why is experiment tracking the foundation of reproducibility?

MLflow is an open source standard for tracking. It logs parameters, metrics, artifacts (models, graphs), and code. MLflow Model Registry is a centralized model storage with versioning and lifecycle stages (Staging → Production → Archived). Deployment via MLflow Serving or integration with external systems.

Typical initialization in code:

import mlflow

mlflow.set_experiment("fraud-detection-v2")
with mlflow.start_run():
    mlflow.log_params({"learning_rate": 3e-4, "batch_size": 64, "epochs": 10})
    mlflow.log_metric("val_f1", val_f1, step=epoch)
    mlflow.pytorch.log_model(model, "model")

This is the minimum. In production, we add logging of system metrics (GPU utilization, memory), dataset (hash, version), code (git commit hash). Weights & Biases — richer UI, collaboration features, sweep for hyperparameter optimization. MLflow — for on-premise deployment without external dependencies.

DVC (Data Version Control) — versioning of data and models on top of git. Data is stored in S3/GCS/Azure Blob, only metadata (hashes) in git. dvc repro reproduces the entire pipeline from raw data to metrics.

To ensure reproducibility of training, fix random seeds (torch.manual_seed, numpy.random.seed, random.seed) and record them in experiment metadata. Without this, debugging irregular results is painful. Log the dataset version (DVC hash) and git commit — then any experiment can be reproduced down to the byte.

Pipeline Orchestration: Kubeflow, Airflow, Prefect

A pipeline orchestrator becomes necessary when: A 100-line training script in cron is fine for simple tasks. But as soon as you have a multi-step pipeline (data loading → preprocessing → feature engineering → training → validation → deployment if quality above threshold), you need an orchestrator with retry logic, visualization, and alerts.

Kubeflow — Kubernetes-native orchestrator for ML (see Kubeflow). Each step is a Docker container. Supports parallel steps, conditional branches, artifacts between steps. Integrates with Katib (AutoML), KServe (serving), Feast (feature store).

Apache Airflow — more general DAG orchestrator. Wide ecosystem of operators (S3, Spark, DBT, Kubernetes). Easier to deploy if Airflow already exists in the company.

Prefect / Metaflow — less boilerplate. Prefect 2.x with @flow and @task decorators — quick start for small teams.

Typical training pipeline architecture on Kubeflow:

1. Data ingestion component — fetches data from S3/DB, validates schema via Great Expectations
2. Preprocessing component — transformations, normalization, train/val/test split
3. Training component — training on GPU, logging to MLflow
4. Evaluation component — metric calculation, comparison with baseline in Model Registry
5. Conditional deployment — deploy only if new model is better than current by >2% F1

Each component is a separate Docker image. Pipeline is versioned in git. Scheduled run (retraining once a week on new data) or manual.

Model Registry and Lifecycle Management

Model Registry is not just a checkpoint store. It is a centralized system that knows:

• Which model is currently in production (and with what metrics)
• History of all versions with training parameters
• Metadata: dataset, git commit, validation results
• Lifecycle stage: None → Staging → Production → Archived

MLflow Model Registry — standard. For enterprise — Vertex AI Model Registry (GCP), SageMaker Model Registry (AWS), Azure ML Model Registry.

Model promotion through stages: automatically move model to Staging after successful eval, then manual or automatic (during A/B test) promotion to Production. Rollback — switch to previous Production version in seconds.

Serving: From FastAPI to Triton Inference Server

Simple case. FastAPI + PyTorch/ONNX on one server — 80% of production ML deployments are exactly that. Sufficient for most tasks with load up to 100 req/s.

from fastapi import FastAPI
import onnxruntime as ort

app = FastAPI()
session = ort.InferenceSession("model.onnx", providers=["CUDAExecutionProvider"])

@app.post("/predict")
async def predict(request: PredictRequest):
    inputs = preprocess(request.text)
    outputs = session.run(None, {"input_ids": inputs})
    return {"label": postprocess(outputs)}

Triton Inference Server — production standard for high loads (500+ req/s). Dynamic batching, concurrent model execution, model ensemble. Supports TensorRT, ONNX, PyTorch TorchScript, TensorFlow SavedModel.

KServe — Kubernetes-native ML serving with autoscaling, canary deployments, A/B testing out of the box. Scale-to-zero for inactive models — savings on infrastructure up to 40% annually for a project with 10 models.

Monitoring: Data Drift, Model Drift, Infrastructure Metrics

Monitoring — what is usually done last and regretted first. Three levels.

Infrastructure monitoring. Latency (P50/P95/P99), throughput (req/s), error rate (4xx, 5xx), GPU/CPU utilization. Prometheus + Grafana — standard. Alert when P99 latency > threshold or error rate > 1%.

Data drift monitoring. Distribution of input data changes over time. Detect via PSI (Population Stability Index) for numerical features: PSI > 0.2 — strong drift. Chi-squared test for categorical, Kolmogorov-Smirnov test for continuous. Evidently AI — open source library with ready-made drift tests.

Model drift monitoring. If ground truth is delayed (e.g., we know conversion after a week) — monitor real metrics. If not — surrogate metrics: distribution of prediction scores, proportion of confident predictions.

Alerting. Three levels: INFO (minor drift, log it), WARNING (significant, notify team), CRITICAL (quality dropped below threshold — automatic switch to fallback model).

Why is data drift monitoring important?

Without it, you learn about model degradation only from user complaints or ringing SLA. A drift alert allows you to retrain the model in advance, before errors start causing losses. In one of our projects, PSI monitoring detected drift 2 days after a data source change — this saved the campaign.

Feature Store

Feature Store solves the training-serving skew problem. If preprocessing during training and inference is implemented in two different places — divergence is inevitable.

A Feature Store is needed when:

• Several models use the same features
• Features are computed from streaming data (real-time)
• Large team with different people on feature engineering and model training

Feast — open source Feature Store. Offline store (S3 + Parquet) for training, online store (Redis, DynamoDB) for low-latency inference. Feature definitions as code, materialization job syncs offline → online.

Tecton (commercial), Vertex AI Feature Store (GCP), SageMaker Feature Store (AWS) — managed options with less ops overhead.

CI/CD for ML

ML CI/CD is regular CI/CD plus specific ML steps.

ML-specific checks in CI:

• Reproducibility check: run training with a fixed seed, result must match
• Data validation: Great Expectations or Pandera on schema/distribution checks
• Model performance check: automatic eval on holdout, block merge if degradation > threshold
• Latency regression test: inference must meet SLA

GitOps for deployment. Merge to main → CI triggers training → eval → if passes → automatic deployment to Staging → smoke tests → manual promotion to Production or automatic upon successful canary.

Tools: GitHub Actions / GitLab CI for CI, ArgoCD for GitOps deployment on Kubernetes.

What's Included in MLOps Platform Development

We provide a full cycle of work, documentation, and team training.

Total time from audit to full MLOps platform: 3–5 months. Also possible phased launch: basic level (tracking + serving) in 4–6 weeks.

Cost is calculated individually based on data volume, number of models, and infrastructure requirements. Order an MLOps infrastructure audit — get a roadmap in 1–2 weeks. Contact us for a project assessment — we will send a preliminary estimate within 2 business days.

Note: warranty on architectural solutions — 12 months. We provide integration certificates with major cloud providers (AWS, GCP, Azure). During our work, we have not lost a single client after the first implementation — the experience of 50+ successful MLOps projects speaks for itself. Get a consultation on building an MLOps platform today.