Custom Prompt Registry Development: Versioning, API & SSO Integration

Note: when the number of prompts in LLM systems exceeds a dozen, chaos begins. We once saw a project where versions were stored in Jira, Confluence, Slack, and even in code comments. A typical typo in a prompt — and a $2500 payment request went out with incorrect data. Deploying a change to 50 microservices took 3 days of manual copying. A Prompt Registry is a prompt management system that centralizes storage, versioning, and deployment. We develop custom Prompt Registry turnkey for companies that find ready-made solutions (PromptLayer, Humanloop) insufficient. Our experience — 10+ years in AI/ML, 50+ launched projects, including RegTech and FinTech with strict security requirements. Our certified engineers ensure a seamless deployment and provide a guaranteed uptime SLA.

According to OpenAI developers, prompt versioning is a key element of production-ready LLM systems.

Problems that a centralized prompt hub solves

Without a unified prompt registry, three critical problems eventually arise:

• Lack of versioning. During audits — compliance failure. No one knows which prompt was used for each request. Restoring history is a manual search through logs and chats.
• Manual propagation of changes. Updating a prompt across 50 microservices is a nightmare for DevOps. With a registry — one API request, and all clients pick up the new version in seconds. Rollback — one call.
• No quality monitoring. We implement metric collection: latency p99, tokens, request cost, quality score. This enables A/B testing of prompts and selecting the best one. Typical quality improvement — 15–30%.

Why do companies need a custom prompt registry?

Ready-made services (PromptLayer, Humanloop) are a good entry-level, but they do not meet corporate needs. Here are key differences:

The table below shows real improvements after custom registry implementation:

A custom Prompt Registry pays for itself by reducing deployment time by 10x compared to manual management. Typical MLOps budget savings — 40%, which for a mid-size company translates to $50,000–$100,000 annually. Engineers' time is freed for more valuable tasks. The solution is suitable for LLM prompt management in large companies with high security requirements.

How do we ensure versioning and rollbacks?

The data schema is based on PostgreSQL. Each prompt version is protected by a SHA256 hash — duplicates are eliminated. On deployment, a record is created in prompt_deployments with a reference to the previous version (rollback_of). Rollback — one API call.

-- PostgreSQL schema for our custom prompt registry
CREATE TABLE prompts (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    name VARCHAR(255) NOT NULL,
    description TEXT,
    created_at TIMESTAMPTZ DEFAULT NOW(),
    created_by VARCHAR(255) NOT NULL,
    tags TEXT[]
);

CREATE TABLE prompt_versions (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    prompt_id UUID REFERENCES prompts(id),
    version_number INTEGER NOT NULL,
    content TEXT NOT NULL,
    content_hash VARCHAR(64) NOT NULL,  -- SHA256
    model VARCHAR(100) NOT NULL,
    temperature FLOAT DEFAULT 0.0,
    max_tokens INTEGER DEFAULT 1000,
    variables JSONB DEFAULT '[]',
    metadata JSONB DEFAULT '{}',
    created_at TIMESTAMPTZ DEFAULT NOW(),
    created_by VARCHAR(255) NOT NULL,
    UNIQUE(prompt_id, version_number)
);

CREATE TABLE prompt_deployments (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    prompt_version_id UUID REFERENCES prompt_versions(id),
    environment VARCHAR(50) NOT NULL,  -- dev/staging/production
    deployed_at TIMESTAMPTZ DEFAULT NOW(),
    deployed_by VARCHAR(255) NOT NULL,
    is_active BOOLEAN DEFAULT TRUE,
    rollback_of UUID  -- Reference to previous version on rollback
);

CREATE TABLE prompt_executions (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    prompt_version_id UUID REFERENCES prompt_versions(id),
    executed_at TIMESTAMPTZ DEFAULT NOW(),
    input_variables JSONB,
    rendered_prompt TEXT,
    response TEXT,
    input_tokens INTEGER,
    output_tokens INTEGER,
    latency_ms INTEGER,
    cost_usd FLOAT,
    quality_score FLOAT  -- Quality score (if available)
);

Fast API on FastAPI

We use FastAPI to create the prompt registry, ensuring high performance. Example of creating a version and retrieving the latest active one:

from fastapi import FastAPI, HTTPException, Depends
from pydantic import BaseModel
import asyncpg

app = FastAPI(title="Prompt Registry API")

class PromptCreateRequest(BaseModel):
    name: str
    content: str
    model: str = "gpt-4o"
    temperature: float = 0.0
    description: str = None

@app.post("/prompts/{name}/versions")
async def create_version(
    name: str,
    request: PromptCreateRequest,
    db = Depends(get_db)
):
    content_hash = hashlib.sha256(request.content.encode()).hexdigest()
    existing = await db.fetchrow(
        "SELECT id FROM prompt_versions pv JOIN prompts p ON p.id = pv.prompt_id "
        "WHERE p.name = $1 AND pv.content_hash = $2",
        name, content_hash
    )
    if existing:
        raise HTTPException(400, "Identical prompt version already exists")

    version = await db.fetchrow("""
        INSERT INTO prompt_versions (prompt_id, version_number, content,
            content_hash, model, temperature)
        SELECT p.id,
               COALESCE(MAX(pv.version_number), 0) + 1,
               $2, $3, $4, $5
        FROM prompts p
        LEFT JOIN prompt_versions pv ON pv.prompt_id = p.id
        WHERE p.name = $1
        GROUP BY p.id
        RETURNING id, version_number
    """, name, request.content, content_hash, request.model, request.temperature)

    return {"version_id": str(version['id']), "version": version['version_number']}

@app.get("/prompts/{name}/latest")
async def get_latest(name: str, environment: str = "production", db = Depends(get_db)):
    prompt = await db.fetchrow("""
        SELECT pv.content, pv.model, pv.temperature, pv.variables, pv.version_number
        FROM prompt_versions pv
        JOIN prompt_deployments pd ON pd.prompt_version_id = pv.id
        JOIN prompts p ON p.id = pv.prompt_id
        WHERE p.name = $1 AND pd.environment = $2 AND pd.is_active = TRUE
        ORDER BY pd.deployed_at DESC LIMIT 1
    """, name, environment)

    if not prompt:
        raise HTTPException(404, f"No deployed prompt '{name}' in {environment}")
    return dict(prompt)

Python client for integration

For ease of use, we write a client library in Python. It caches the latest active version and substitutes template variables:

class PromptClient:
    def __init__(self, registry_url: str, api_key: str):
        self.url = registry_url
        self.headers = {"X-API-Key": api_key}
        self._cache = {}

    def get_and_render(self, name: str, variables: dict,
                       environment: str = "production") -> str:
        cache_key = f"{name}:{environment}"
        if cache_key not in self._cache:
            resp = requests.get(
                f"{self.url}/prompts/{name}/latest",
                params={"environment": environment},
                headers=self.headers
            )
            self._cache[cache_key] = resp.json()

        template = self._cache[cache_key]['content']
        for var, value in variables.items():
            template = template.replace(f"{{{{{var}}}}}", str(value))
        return template

Prompt versioning for compliance

Full change history with SHA256 hash and metadata allows delivering an audit report in minutes: which prompt, when, and by whom was used. Our RegTech clients reduce audit time from weeks to hours. A custom solution implements changes 5x faster than manual microservice updates. SSO integration for the prompt registry is a standard option that simplifies the audit trail.

Work process

1. Analysis — audit of current practices, authentication requirements, compliance, volumes. Define target metrics (p99 latency < 30 ms, throughput > 1000 rps).
2. Design — database schema, API architecture, deployment model (Kubernetes, bare-metal).
3. Implementation — backend development, client, CI/CD integration (GitLab CI, GitHub Actions).
4. Testing — unit, integration, load tests (p99 latency measured up to 50 ms).
5. Deployment — rollout on your environments, documentation, team training.
6. Support — SLA, monitoring, improvements based on feedback.

Our engineers are proficient in prompt engineering and can tailor the system to any business processes. With over 5 years of MLOps experience and 50+ completed projects, we deliver a proven solution.

Timeline and what's included

Approximately — from 4 to 8 weeks depending on integration complexity (SSO, custom metrics). Includes: working system, documentation (API + administration), training for up to 5 people, source code, 1 month support. Cost is calculated individually — contact us, we will evaluate your project. Typical projects cost $30k–$80k.

Typical result after implementation: deployment speed of changes increases 10x, deployment error rate decreases by 95%, audit time shrinks from weeks to minutes. A unified Prompt Registry becomes a key component of the MLOps infrastructure.

Get a consultation: tell us about your tasks — we will propose a solution considering your data governance and budget. Order a preliminary analysis through the form on the website.

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.