Cost Tracking for AI Requests: Monitor Token Costs

Setting Up Cost Tracking for AI Requests

Imagine: Monday morning you see an OpenAI bill for $15,000, though you planned $10,000 a month. Sound familiar? Without tracking token costs by project and user, such overspend is only a matter of time. Cost tracking for LLMs is not an option but a necessity if you want to manage your AI infrastructure budget. Our experience shows: companies lose up to 40% of their budget on uncontrolled requests. We'll set up a system that calculates costs in real time for each model, project, and feature. As a result, you'll be able to plan expenses accurately and detect anomalies early.

What Problems Does Cost Tracking Solve?

Without an accounting system, you won't see which department or feature generates 80% of costs. A typical case: a RAG feature for customer support uses long contexts — 50% of all tokens go to it, but that's invisible until the bill arrives. Another problem is uncontrolled tests in dev environments: developers run heavy prompts on GPT-4o, though gpt-4o-mini would suffice for testing. Another pain point is the lack of alerts: if costs surge tenfold in a day due to a bug, you only learn about it at the end of the month. Our solution provides transparency: you see costs in real time, and when thresholds are exceeded — instant notification.

To understand the root cause, let's break down typical setup mistakes. First, many don't log the token count per request — without that, accurate cost calculation is impossible. Second, they ignore aggregation by feature: they only see the total bill, not what exactly is expensive. Third, they don't set budget limits — then any code error (e.g., an infinite loop calling the LLM) can burn through the monthly budget in an hour.

How We Set Up Token Accounting?

Each request to an LLM generates input and output tokens. The cost depends on the model and the number of tokens. For example, for GPT-4o the price per 1M input tokens is $2.50, output $10.00. For a local LLaMA 3-8B, costs are determined solely by GPU resources.

# Example pricing (update periodically via API)
MODEL_PRICING = {
    "gpt-4o": {"input": 2.50, "output": 10.00},
    "gpt-4o-mini": {"input": 0.15, "output": 0.60},
    "claude-3-5-sonnet-20241022": {"input": 3.00, "output": 15.00},
    "claude-3-haiku-20240307": {"input": 0.25, "output": 1.25},
    "llama-3-8b-local": {"input": 0.0, "output": 0.0},
}

def calculate_cost(model: str, prompt_tokens: int, completion_tokens: int) -> float:
    if model not in MODEL_PRICING:
        return 0.0
    pricing = MODEL_PRICING[model]
    return (prompt_tokens * pricing["input"] + completion_tokens * pricing["output"]) / 1_000_000

Aggregation by Dimensions

For analysis, we aggregate cost across multiple dimensions: project, user, feature tag (chat, RAG, classification), time interval. This allows answering the question: "Why did the AI bill increase by 30% in a week?"

class CostTracker:
    def record(self, request: LLMRequest, cost_usd: float):
        self.db.insert({
            "timestamp": request.timestamp,
            "cost_usd": cost_usd,
            "model": request.model,
            "project_id": request.project_id,
            "user_id": request.user_id,
            "feature": request.feature_tag,
            "prompt_tokens": request.prompt_tokens,
            "completion_tokens": request.completion_tokens,
        })

    def get_daily_by_project(self, days: int = 30) -> dict:
        return self.db.query("""
            SELECT project_id, DATE(timestamp) as date,
                   SUM(cost_usd) as total_cost,
                   SUM(prompt_tokens) as total_tokens
            FROM llm_costs
            WHERE timestamp > NOW() - INTERVAL %s DAY
            GROUP BY project_id, date
            ORDER BY date, total_cost DESC
        """, (days,))

What Alerts Help Stay Within Budget?

Budget alerts are a key element of Cost Tracking. We configure two thresholds: daily and hourly. When 80% of the limit is reached, a warning is sent to Slack or Telegram; at 100%, automatic throttling of requests is triggered.

class BudgetGuard:
    def __init__(self, limits: dict):
        self.daily_limit_usd = limits["daily"]
        self.hourly_limit_usd = limits["hourly"]

    def check_budget(self, project_id: str) -> BudgetStatus:
        daily_spend = self.tracker.get_spend(project_id, hours=24)
        hourly_spend = self.tracker.get_spend(project_id, hours=1)

        alerts = []
        if daily_spend > self.daily_limit_usd * 0.8:
            alerts.append(f"80% of daily budget consumed: ${daily_spend:.2f}/${self.daily_limit_usd:.2f}")
        if daily_spend > self.daily_limit_usd:
            alerts.append("DAILY BUDGET EXCEEDED — throttling enabled")

        return BudgetStatus(daily_spend=daily_spend, alerts=alerts,
                           throttle_enabled=daily_spend > self.daily_limit_usd)

Why Real-Time Monitoring Is Better?

Even an hour delay in accounting can lead to overspending. For example, launching a new feature without limits can cause costs to skyrocket tenfold in a day. A system with alerts reacts in minutes, not days. Our solution saves up to 30% of the budget through early anomaly detection.

What's Included in Cost Tracking Setup?

We provide:

• Integration with provider logs (OpenAI, Anthropic, local models)
• Dashboard with charts (daily cost, top features, cost per request)
• Alert configuration in messengers
• Documentation for adding new models
• Support for one month after deployment

Implementation timeline: 5 to 15 days depending on infrastructure complexity. Cost is calculated individually — contact us for a project assessment. Get a consultation on AI cost optimization.

Comparison: Custom Tracker vs Our Solution

Our solution reduces implementation time by 2-4 times compared to a custom tracker. Our team has over 5 years in ML infrastructure and 50+ Cost Tracking implementations for AI products. Contact us for a consultation.

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.