Prompt Versioning System for LLMs

In one support chatbot project, 15 prompt variants accumulated over a month. No one remembered what had changed or which metrics deteriorated. Rolling back to a working version took up to 3 hours. After implementing our system, rollback time dropped to 2 minutes, and regression frequency fell by 40%. We developed a prompt versioning system that gives full control over changes, automatically links each version to metrics (ROUGE-L, BLEU, human rating), and prevents accidental regressions. Our extensive experience allows us to manage prompts effectively in teams of any size.

What problems does prompt versioning solve?

Loss of change history. Without a versioning system, developers edit prompts directly in production, losing context. A week later, no one can explain why the model's behavior changed.

Unnoticed regressions. Changing a single phrase can drop answer accuracy by 10–15%. Without tying to metrics, such drops go unnoticed until user complaints.

Long search for a working version. When metrics drop, the team spends hours searching old versions in chats and files. Our system stores every version immutably and allows rollback in minutes.

Lack of A/B testing. Without versioning, it is impossible to run parallel tests of different prompts in a controlled environment. We add A/B testing capability with gradual rollout.

How does semantic prompt versioning work?

We use the Semantic Versioning standard (major.minor.patch):

• Major — change of task or request architecture (e.g., switching models or adding a new data type).
• Minor — improvement of wording, addition of few-shot examples, change of tone.
• Patch — fixing typos, minor edits.

Each version is tied to metrics on the evaluation set: ROUGE-L, BLEU, human rating 1–5, latency p99. The regression threshold is 3%: if a metric drops more than that, CI blocks promotion.

Git-based prompt storage

For small teams, storing prompts in Git with accompanying YAML files is sufficient.

Example structure:

prompts/
├── customer-support/
│   ├── system-prompt.v1.txt
│   ├── system-prompt.v2.txt
│   └── system-prompt.current -> system-prompt.v2.txt
├── summarization/
│   ├── prompt.v1.yaml
│   └── prompt.v2.yaml
└── prompts.json  # index with metadata

The YAML file contains version, author, changelog, model, variables, prompt text, and metrics.

# prompts/summarization/prompt.v2.yaml
version: "2.0.0"
name: "document-summarizer"
author: "ml-team"
changelog: "Added length constraint, improved tone instruction"
model:
  provider: "openai"
  name: "gpt-4o"
  temperature: 0.2
  max_tokens: 500
variables:
  - name: document
    required: true
  - name: max_sentences
    required: false
    default: "3"
content: |
  Summarize the following document in exactly {{max_sentences}} sentences.
  Be concise and focus on the main points.
  Do not add information not present in the document.

  Document:
  {{document}}
metrics:
  rouge_l: 0.47
  human_rating: 4.2
  eval_set: "summarization-benchmark-v3"

Programmatic diff of prompts

import difflib

def diff_prompt_versions(v1_content: str, v2_content: str) -> str:
    v1_lines = v1_content.splitlines(keepends=True)
    v2_lines = v2_content.splitlines(keepends=True)
    diff = difflib.unified_diff(
        v1_lines, v2_lines,
        fromfile="version_1",
        tofile="version_2",
        lineterm=""
    )
    return "".join(diff)

def analyze_prompt_change(v1: str, v2: str) -> dict:
    v1_words = set(v1.lower().split())
    v2_words = set(v2.lower().split())
    added_words = v2_words - v1_words
    removed_words = v1_words - v2_words
    return {
        "length_change": len(v2) - len(v1),
        "added_words": list(added_words)[:10],
        "removed_words": list(removed_words)[:10],
        "similarity": difflib.SequenceMatcher(None, v1, v2).ratio(),
        "change_type": "major" if difflib.SequenceMatcher(None, v1, v2).ratio() < 0.7 else "minor"
    }

How to automatically rollback a prompt on regression?

When metric drops exceed 3%, the pipeline blocks promotion. The developer sees the diff and the list of affected metrics. In a critical situation, you can manually switch the symlink to the previous version — this takes seconds. In one case, we rolled back a sales chatbot prompt in 2 minutes, restoring conversion to its previous level.

Comparison with manual process:

Why is prompt versioning important?

Prompt versioning is a basic MLOps element for LLMs. Without it, any change in production can lead to unexpected model behavior that is difficult to roll back. A versioning system provides experiment reproducibility and confidence that every prompt has been tested on an evaluation set before deployment. This is especially critical for customer-facing applications where the cost of error is high.

Prompt promotion process

The process consists of stages:

[Draft] → [In Review] → [Approved] → [Staging] → [Production]
                ↑                         ↓
           Reviewer                  A/B Test (5%)
                                          ↓
                                    Full Rollout / Rollback

Key rule: no prompts go into production without passing the evaluation set. An automated CI job runs on each change and blocks promotion if regression exceeds 3%.

Checklist for implementing a versioning system

1. Conduct an audit of current prompts — collect all versions, document metrics.
2. Choose storage method: Git (for small teams) or specialized platform (LangSmith, custom backend).
3. Set up an evaluation set — at least 100–500 examples for reliable assessment.
4. Integrate CI/CD: automated tests on every push to the prompt repository.
5. Define regression thresholds for blocking promotion (we recommend 3–5%).
6. Train the team to work with the system: creating versions, reviewing, rolling back.

What our work includes

• Audit of the current prompt management pipeline.
• Designing a versioning schema: data model selection, integration with Git or specialized storage.
• Backend development, linking to evaluation set, setting up CI/CD.
• API documentation, version schemas, team instructions.
• Workshop for developers: how to create, review, and roll back prompts.
• Two weeks of free support after deployment.

Our team has extensive experience in ML and NLP, having implemented versioning systems for three large projects with dozens of prompts. Order an audit of your prompts and get recommendations for versioning. Contact us for a free consultation — we will analyze your pipeline and offer the optimal solution.

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.