Engineering MLMLOpsLLM

Senior ML Engineer

ML model deployment, MLOps pipelines, LLM integration, and cost optimization.

$skillfish add ~skills/senior-ml-engineer

Real Examples Get Started Models & Context

Real Conversations

See how the Senior ML Engineer skill responds to actual requests. These are representative examples of what your AI agent will produce.

You

Fine-tuned LLaMA 3.1 8B with QLoRA on our dataset. Need to serve it with <200ms p95 latency on a single A10G GPU. What's the best serving stack?

Agent

Serving Stack: vLLM on A10G with quantization

vLLM's PagedAttention gives 2-4x higher throughput vs naive HuggingFace inference, and easily fits 8B in 24GB VRAM with 4-bit quantization.

Merge LoRA weights first:

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained('meta-llama/Meta-Llama-3.1-8B')
model = PeftModel.from_pretrained(base, './lora-checkpoint')
merged = model.merge_and_unload()
merged.save_pretrained('./llama3-finetuned-merged')

vLLM server:

pip install vllm
python -m vllm.entrypoints.openai.api_server \
  --model ./llama3-finetuned-merged \
  --quantization awq \
  --max-model-len 4096 \
  --gpu-memory-utilization 0.90 \
  --tensor-parallel-size 1 \
  --port 8000

AWQ quantization (do this once before deploying):

pip install autoawq
autoawq quantize ./llama3-finetuned-merged --quant_config '{"zero_point": true, "q_group_size": 128, "w_bit": 4}'

Expected performance on A10G (24GB):

Config	VRAM	p50 latency	p95 latency
FP16, no batching	18GB	80ms	180ms
AWQ 4-bit + vLLM	10GB	45ms	120ms

Monitoring: Track vllm:num_requests_running and vllm:gpu_cache_usage_perc in Prometheus — evict when cache usage >85%.

You

Have 5 data scientists running experiments with different sklearn models. No tracking — results get lost in Jupyter notebooks. Need MLflow set up properly.

Agent

MLflow Setup for Team Experiment Tracking:

1. Shared MLflow server (deploy on internal VM or ECS):

mlflow server \
  --backend-store-uri postgresql://mlflow:password@db-host/mlflow \
  --default-artifact-root s3://your-bucket/mlflow-artifacts \
  --host 0.0.0.0 --port 5000

2. Standardized experiment decorator for your team:

import mlflow
from sklearn.pipeline import Pipeline
from sklearn.model_selection import cross_val_score
import numpy as np

MLFLOW_URI = 'http://mlflow-server:5000'

def run_experiment(experiment_name: str, pipeline: Pipeline, X, y, params: dict):
    mlflow.set_tracking_uri(MLFLOW_URI)
    mlflow.set_experiment(experiment_name)

    with mlflow.start_run():
        mlflow.log_params(params)
        mlflow.log_param('model_type', type(pipeline.steps[-1][1]).__name__)

        scores = cross_val_score(pipeline, X, y, cv=5, scoring='roc_auc')
        mlflow.log_metric('cv_auc_mean', np.mean(scores))
        mlflow.log_metric('cv_auc_std', np.std(scores))

        pipeline.fit(X, y)
        mlflow.sklearn.log_model(pipeline, 'model',
            registered_model_name=experiment_name,
            input_example=X[:5])

        print(f'Run logged: {mlflow.active_run().info.run_id}')

3. Usage by any data scientist:

run_experiment(
    experiment_name='churn-prediction',
    pipeline=Pipeline([('scaler', StandardScaler()), ('clf', XGBClassifier(n_estimators=200))]),
    X=X_train, y=y_train,
    params={'n_estimators': 200, 'feature_set': 'v3', 'dataset_date': '2025-03'}
)

Model promotion: UI at :5000 — compare runs, promote best to Staging → Production with one click.

Get Started

How to install and use this skill in your preferred environment.

Skills are designed for AI coding agents (Claude Code, Cursor, Windsurf) and IDE-based workflows where the agent can read files, run scripts, and act on your codebase. Web-based AI can use the knowledge and frameworks, but won't have tool access.

Models & Context

Which AI models and context windows work best with this skill.

Recommended Models

Best

Claude Opus 4 Claude Sonnet 4 GPT-4.1 Gemini 2.5 Pro Grok 3 Kimi K2

Good

Claude Haiku 4.5 GPT-4.1 mini Gemini 2.5 Flash Grok 3 mini

Larger models produce more detailed, production-ready outputs.

Context Window

This skill's SKILL.md is typically 3–10 KB — fits in any modern context window.

8K Skill only

32K+ Skill + conversation

100K+ Skill + references + codebase

All current frontier models (Claude, GPT, Gemini) support 100K+ context. Use the full window for complex multi-service work.

Pro tips for best results

1

Be specific

Include numbers — users, budget, RPS — so the skill can size the architecture.

2

Share constraints

Compliance needs, team size, and existing stack all improve the output.

3

Iterate

Start with a high-level design, then ask follow-ups for IaC, cost analysis, or security review.

4

Combine skills

Pair with companion skills below for end-to-end coverage.

Ready to try Senior ML Engineer?

Install the skill and start getting expert-level guidance in your workflow — any agent, any IDE.

$skillfish add ~skills/senior-ml-engineer

← Browse all 169 skills