senior-ml-engineer скилл для Claude Code: что делает и как ставить

---
name: "senior-ml-engineer"
description: ML engineering skill for productionizing models, building MLOps pipelines, and integrating LLMs. Covers model deployment, feature stores, drift monitoring, RAG systems, and cost optimization. Use when the user asks about deploying ML models to production, setting up MLOps infrastructure (MLflow, Kubeflow, Kubernetes, Docker), monitoring model performance or drift, building RAG pipelines, or integrating LLM APIs with retry logic and cost controls. Focused on production and operational concerns rather than model research or initial training.
metadata:
  triggers:
    - MLOps pipeline
    - model deployment
    - feature store
    - model monitoring
    - drift detection
    - RAG system
    - LLM integration
    - model serving
    - A/B testing ML
    - automated retraining
---

# Senior ML Engineer

Production ML engineering patterns for model deployment, MLOps infrastructure, and LLM integration.

---

## Table of Contents

- [Model Deployment Workflow](#model-deployment-workflow)
- [MLOps Pipeline Setup](#mlops-pipeline-setup)
- [LLM Integration Workflow](#llm-integration-workflow)
- [RAG System Implementation](#rag-system-implementation)
- [Model Monitoring](#model-monitoring)
- [Reference Documentation](#reference-documentation)
- [Tools](#tools)

---

## Model Deployment Workflow

Deploy a trained model to production with monitoring:

1. Export model to standardized format (ONNX, TorchScript, SavedModel)
2. Package model with dependencies in Docker container
3. Deploy to staging environment
4. Run integration tests against staging
5. Deploy canary (5% traffic) to production
6. Monitor latency and error rates for 1 hour
7. Promote to full production if metrics pass
8. **Validation:** p95 latency < 100ms, error rate < 0.1%

### Container Template

```dockerfile
FROM python:3.11-slim

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

COPY model/ /app/model/
COPY src/ /app/src/

HEALTHCHECK CMD curl -f http://localhost:8080/health || exit 1

EXPOSE 8080
CMD ["uvicorn", "src.server:app", "--host", "0.0.0.0", "--port", "8080"]
```

### Serving Options

| Option | Latency | Throughput | Use Case |
|--------|---------|------------|----------|
| FastAPI + Uvicorn | Low | Medium | REST APIs, small models |
| Triton Inference Server | Very Low | Very High | GPU inference, batching |
| TensorFlow Serving | Low | High | TensorFlow models |
| TorchServe | Low | High | PyTorch models |
| Ray Serve | Medium | High | Complex pipelines, multi-model |

---

## MLOps Pipeline Setup

Establish automated training and deployment:

1. Configure feature store (Feast, Tecton) for training data
2. Set up experiment tracking (MLflow, Weights & Biases)
3. Create training pipeline with hyperparameter logging
4. Register model in model registry with version metadata
5. Configure staging deployment triggered by registry events
6. Set up A/B testing infrastructure for model comparison
7. Enable drift monitoring with alerting
8. **Validation:** New models automatically evaluated against baseline

### Feature Store Pattern

```python
from feast import Entity, Feature, FeatureView, FileSource

user = Entity(name="user_id", value_type=ValueType.INT64)

user_features = FeatureView(
    name="user_features",
    entities=["user_id"],
    ttl=timedelta(days=1),
    features=[
        Feature(name="purchase_count_30d", dtype=ValueType.INT64),
        Feature(name="avg_order_value", dtype=ValueType.FLOAT),
    ],
    online=True,
    source=FileSource(path="data/user_features.parquet"),
)
```

### Retraining Triggers

| Trigger | Detection | Action |
|---------|-----------|--------|
| Scheduled | Cron (weekly/monthly) | Full retrain |
| Performance drop | Accuracy < threshold | Immediate retrain |
| Data drift | PSI > 0.2 | Evaluate, then retrain |
| New data volume | X new samples | Incremental update |

---

## LLM Integration Workflow

Integrate LLM APIs into production applications:

1. Create provider abstraction layer for vendor flexibility
2. Implement retry logic with exponential backoff
3. Configure fallback to secondary provider
4. Set up token counting and context truncation
5. Add response caching for repeated queries
6. Implement cost tracking per request
7. Add structured output validation with Pydantic
8. **Validation:** Response parses correctly, cost within budget

### Provider Abstraction

```python
from abc import ABC, abstractmethod
from tenacity import retry, stop_after_attempt, wait_exponential

class LLMProvider(ABC):
    @abstractmethod
    def complete(self, prompt: str, **kwargs) -> str:
        pass

@retry(stop=stop_after_attempt(3), wait=wait_exponential(min=1, max=10))
def call_llm_with_retry(provider: LLMProvider, prompt: str) -> str:
    return provider.complete(prompt)
```

### Cost Management

Do not hardcode prices, and do not trust a price table you find in a document
(including this one). Providers reprice several times a year, and a stale
figure produces a confidently wrong business case.

Work in tiers and look the current numbers up at request time:

| Tier | Typical use | Relative cost |
|------|-------------|---------------|
| Small | Classification, extraction, routing, short output | 1x baseline |
| Mid | Summarisation, structured output, moderate reasoning | ~10-25x small |
| Large | Multi-step reasoning, code generation, long context | ~50-100x small |

Read the live rate from your provider's pricing page and pass it in, the way
`engineering-team/skills/senior-prompt-engineer/scripts/prompt_optimizer.py`
takes `--price-per-mtok`.
The ratios between tiers are far more stable than the absolute prices, so
build the model-routing decision on the ratio.

---

## RAG System Implementation

Build retrieval-augmented generation pipeline:

1. Choose vector database (Pinecone, Qdrant, Weaviate)
2. Select embedding model based on quality/cost tradeoff
3. Implement document chunking strategy
4. Create ingestion pipeline with metadata extraction
5. Build retrieval with query embedding
6. Add reranking for relevance improvement
7. Format context and send to LLM
8. **Validation:** Response references retrieved context, no hallucinations

### Vector Database Selection

| Database | Hosting | Scale | Latency | Best For |
|----------|---------|-------|---------|----------|
| Pinecone | Managed | High | Low | Production, managed |
| Qdrant | Both | High | Very Low | Performance-critical |
| Weaviate | Both | High | Low | Hybrid search |
| Chroma | Self-hosted | Medium | Low | Prototyping |
| pgvector | Self-hosted | Medium | Medium | Existing Postgres |

### Chunking Strategies

| Strategy | Chunk Size | Overlap | Best For |
|----------|------------|---------|----------|
| Fixed | 500-1000 tokens | 50-100 | General text |
| Sentence | 3-5 sentences | 1 sentence | Structured text |
| Semantic | Variable | Based on meaning | Research papers |
| Recursive | Hierarchical | Parent-child | Long documents |

---

## Model Monitoring

Monitor production models for drift and degradation:

1. Set up latency tracking (p50, p95, p99)
2. Configure error rate alerting
3. Implement input data drift detection
4. Track prediction distribution shifts
5. Log ground truth when available
6. Compare model versions with A/B metrics
7. Set up automated retraining triggers
8. **Validation:** Alerts fire before user-visible degradation

### Drift Detection

```python
from scipy.stats import ks_2samp

def detect_drift(reference, current, threshold=0.05):
    statistic, p_value = ks_2samp(reference, current)
    return {
        "drift_detected": p_value < threshold,
        "ks_statistic": statistic,
        "p_value": p_value
    }
```

### Alert Thresholds

| Metric | Warning | Critical |
|--------|---------|----------|
| p95 latency | > 100ms | > 200ms |
| Error rate | > 0.1% | > 1% |
| PSI (drift) | > 0.1 | > 0.2 |
| Accuracy drop | > 2% | > 5% |

---

## Reference Documentation

### MLOps Production Patterns

`references/mlops_production_patterns.md` contains:

- Model deployment pipeline with Kubernetes manifests
- Feature store architecture with Feast examples
- Model monitoring with drift detection code
- A/B testing infrastructure with traffic splitting
- Automated retraining pipeline with MLflow

### LLM Integration Guide

`references/llm_integration_guide.md` contains:

- Provider abstraction layer pattern
- Retry and fallback strategies with tenacity
- Prompt engineering templates (few-shot, CoT)
- Token optimization with tiktoken
- Cost calculation and tracking

### RAG System Architecture

`references/rag_system_architecture.md` contains:

- RAG pipeline implementation with code
- Vector database comparison and integration
- Chunking strategies (fixed, semantic, recursive)
- Embedding model selection guide
- Hybrid search and reranking patterns

---

## Tools

### Model Deployment Pipeline

```bash
python scripts/model_deployment_pipeline.py --model model.pkl --target staging
```

Generates deployment artifacts: Dockerfile, Kubernetes manifests, health checks.

### RAG System Builder

```bash
python scripts/rag_system_builder.py --config rag_config.yaml --analyze
```

Scaffolds RAG pipeline with vector store integration and retrieval logic.

### ML Monitoring Suite

```bash
python scripts/ml_monitoring_suite.py --config monitoring.yaml --deploy
```

Sets up drift detection, alerting, and performance dashboards.

---

## Tech Stack

| Category | Tools |
|----------|-------|
| ML Frameworks | PyTorch, TensorFlow, Scikit-learn, XGBoost |
| LLM Frameworks | LangChain, LlamaIndex, DSPy |
| MLOps | MLflow, Weights & Biases, Kubeflow |
| Data | Spark, Airflow, dbt, Kafka |
| Deployment | Docker, Kubernetes, Triton |
| Databases | PostgreSQL, BigQuery, Pinecone, Redis |