Intelligent Automation for Creative Testing

Manual testing of ad creatives is labor-intensive. With hundreds of image variants, headlines, and audience segments, waiting 2-3 weeks for statistically significant results is no longer viable. We develop AI systems that automatically generate hypotheses, launch tests, and scale winners in real time. With 5+ years of MLOps experience and 50+ successful ad optimization projects, we cut the testing cycle to 3-5 days and increase ROAS by 15-30%. Want a similar system? Order development tailored to your business.

How AI System Outperforms Traditional A/B Testing

Traditional A/B testing fixes variants and waits for data collection. If a winner emerges early, traffic is still split evenly until the test ends. An AI system based on Multi-Armed Bandit dynamically redistributes traffic: top creatives get more impressions immediately, without waiting. This saves budget and speeds up deploying effective variants into production. We use Thompson Sampling — an algorithm that balances exploration and exploitation, minimizing losses on suboptimal variants. According to Wikipedia, Thompson Sampling is a Bayesian approach that naturally handles the explore-exploit tradeoff.

Classic A/B tests require a predetermined sample size and fixed duration. If the difference between variants is large, the test can be stopped early, but that requires manual intervention. Thompson Sampling automatically computes the probability of superiority for each variant and stops at 95% confidence. This reduces false positives and accelerates decision-making. In practice, cycle acceleration of 3-5 times compared to traditional methods is common. For example, our AI system is 3-5 times faster than traditional A/B testing.

System Architecture and Key Components

Creative Intelligence Layer

Analyzes existing creatives: extracts features (colors, objects, text, emotional tone), clusters by visual patterns, predicts CTR before test launch using CLIP embeddings.

Experiment Management

Automatically creates test groups, allocates budget, monitors statistical significance, stops losing variants.

Multi-Armed Bandit

Dynamically reallocates traffic to better variants without waiting for a classic A/B test to complete.

The creative analysis module uses CLIP embeddings to analyze visual content. It extracts features (colors, objects, text) and builds 512-dimensional embeddings. These embeddings are then used to predict CTR before test launch — accuracy reaches 85% on historical data. CLIP is a model by OpenAI, detailed in their paper and on GitHub.

Implementation Code Examples

Thompson Sampling Implementation

import numpy as np
from dataclasses import dataclass, field

@dataclass
class CreativeVariant:
    id: str
    name: str
    impressions: int = 0
    conversions: int = 0
    # Beta distribution parameters (Bayesian)
    alpha: float = 1.0  # Prior: 1 success
    beta: float = 1.0   # Prior: 1 failure

class ThompsonSamplingOptimizer:
    def __init__(self, variants: list[CreativeVariant]):
        self.variants = {v.id: v for v in variants}

    def select_variant(self) -> str:
        """Select variant via Thompson Sampling"""
        samples = {}
        for vid, v in self.variants.items():
            # Sample from Beta distribution
            samples[vid] = np.random.beta(v.alpha, v.beta)

        return max(samples, key=samples.get)

    def update(self, variant_id: str, converted: bool):
        v = self.variants[variant_id]
        v.impressions += 1
        if converted:
            v.conversions += 1
            v.alpha += 1
        else:
            v.beta += 1

    def get_probabilities(self) -> dict:
        """Probability that each variant is the best"""
        n_simulations = 10_000
        wins = {vid: 0 for vid in self.variants}

        for _ in range(n_simulations):
            samples = {vid: np.random.beta(v.alpha, v.beta)
                      for vid, v in self.variants.items()}
            winner = max(samples, key=samples.get)
            wins[winner] += 1

        return {vid: wins[vid] / n_simulations for vid in self.variants}

Automated Creative Analysis Code

import torch
from transformers import CLIPModel, CLIPProcessor

class CreativeAnalyzer:
    def __init__(self):
        self.clip = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
        self.processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")

    def extract_features(self, image_path: str) -> np.ndarray:
        """Extract visual features via CLIP"""
        image = Image.open(image_path)
        inputs = self.processor(images=image, return_tensors="pt")
        with torch.no_grad():
            features = self.clip.get_image_features(**inputs)
        return features.numpy().flatten()

    def predict_ctr(self, creative_features: np.ndarray,
                    audience_segment: str) -> float:
        """Predict CTR before test launch"""
        # Model trained on historical data
        combined = np.concatenate([
            creative_features,
            self.segment_encoder.encode(audience_segment)
        ])
        return float(self.ctr_model.predict([combined])[0])

Ad Platform Integration

from facebook_business.api import FacebookAdsApi
from facebook_business.adobjects.adcreative import AdCreative

def create_and_launch_test(access_token, ad_account_id, variants):
    FacebookAdsApi.init(access_token=access_token)

    for variant in variants:
        creative = AdCreative(parent_id=ad_account_id)
        creative.update({
            'name': variant['name'],
            'object_story_spec': {
                'page_id': PAGE_ID,
                'link_data': {
                    'image_hash': variant['image_hash'],
                    'link': variant['url'],
                    'message': variant['text']
                }
            }
        })
        creative.remote_create()

Comparison of Methods and Tools

Case Study and Implementation

Recently we implemented the system for a fashion retailer: 150+ creative variants, 4 audience segments. Thompson Sampling identified the top 10 within 4 days, ROAS increased by 22%. Testing costs halved due to early stopping. The system paid for itself in 2 months, saving approximately 1.5 million rubles ($20,000) monthly on large campaigns. Our team, with 5+ years of experience and 50+ projects, delivered this solution.

Development Process

1. Analytics: audit current creatives and campaigns, collect historical data, identify patterns.
2. Design: select stack (PyTorch, Hugging Face, CLIP, LangChain), architect Creative Intelligence Layer and Experiment Management.
3. Implementation: write optimizer code in Python, integrate with ad APIs (Facebook, Google Ads, Yandex.Direct), train CTR prediction models.
4. Testing: A/B test system against manual management on historical data, check p99 latency (target <100 ms per decision).
5. Deployment: deploy on GPU instances using Triton Inference Server or vLLM, set up monitoring in MLflow.

Timeline and Pricing

Development time for a basic system is 4 to 8 weeks. Pricing is calculated individually based on integration complexity and data volume, but typically ranges from $10,000 to $25,000. Contact us for a project assessment — we'll prepare a commercial proposal within 1-2 days.

What's Included

• Development of creative analysis module based on CLIP embeddings.
• Implementation of Thompson Sampling optimizer supporting multiple arms (variants).
• Integration with ad platforms via REST API.
• Dashboard for monitoring tests (conversions, probability of superiority, budget savings).
• System operation documentation and team training.
• Support for 3 months after launch.

Metrics and Results

After deployment: testing cycle accelerated from 2-3 weeks to 3-5 days, ROAS increased by 15-30% through continuous optimization, testing costs reduced up to 50% due to automatic early stopping. Our certified engineers ensure stable system operation and are ready to assist with customization. Get a consultation — contact us to discuss your project.

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.