🧠 MLOps Project Generator

Generate production-ready MLOps project scaffolds — via Web UI or Python CLI.
Scikit-learn · PyTorch · TensorFlow · NLP (Hugging Face) · Multi-cloud · 40+ files per project

This stack supports the full MLOps lifecycle:

Data → Train → Track → Orchestrate → Deploy → Monitor → Improve

---

🎉 What's New (Web UI v1.2.0)

Recent improvements to the Next.js Web UI and TypeScript generator:

Web experience

• Observatory UI — redesigned landing with mission, features, how-it-works, and generator sections
• Mobile-first layout — compact on phones; desktop layout unchanged from sm breakpoint up
• Archive preview — GET /api/preview (partial config, no strict validation) and POST /api/preview?strict=true before download
• File tree preview in the form — see 40+ paths before generating
• Recent projects — browser-local history with quick re-download

Generator (TypeScript v1.2.0)

• Legacy Jinja templates wired — repo templates/ synced into web_ui/templates/ and rendered for classification / regression / timeseries
• Hugging Face NLP — src/nlp/finetune.py, configs/nlp.yaml, Trainer + optional FastAPI inference
• Deployment artifacts — root Dockerfile, docker-compose.yml, Kubernetes manifests (k8s/) when Docker/K8s is selected
• Orchestration files — Airflow DAG (dags/ml_training_dag.py), Kubeflow pipeline stub
• Monitoring scripts — Evidently drift report + custom metrics collector
• Config presets & templates — preset_config and custom_template now affect output (CI, notebooks, minimal/comprehensive/microservice bundles)
• Stack presets — also set preset + template fields when applied
• generation_insights.json — optional recommendations when analytics is enabled

Template fixes (templates/)

• Unified timeseries naming (was mixed with time-series)
• Fixed TimeSeriesModel imports via snippets/model_import.py.j2
• Improved requirements.txt.j2, config.yaml.j2, README/Makefile/.env Jinja templates
• New templates/nlp/ for Hugging Face fine-tuning (CLI + web sync)

---

🚀 Features

🏗️ Core generation

Capability	Web UI	Python CLI
Frameworks	sklearn, PyTorch, TensorFlow	Same
Task types	classification, regression, timeseries, nlp, computer-vision	Same
Stack presets	6 curated stacks (form)	init --preset
ZIP download	Yes	Project folder
Generator runtime	Pure TypeScript (API routes)	Jinja2 + Python

🔬 MLOps integration (generated into ZIP)

• Experiment tracking — MLflow, W&B, custom, or none
• Orchestration — Airflow DAG, Kubeflow pipeline YAML (not empty folders only)
• Deployment — FastAPI inference, Docker (+ compose), Kubernetes (deployment/service/configmap)
• Monitoring — Evidently drift script, custom metrics collector
• Cloud packs — AWS / GCP / Azure folders with Dockerfile, config, deploy scripts

📦 Template profiles (Web UI — optional fields)

Preset config	Effect
quick-start	Lean output (e.g. skips Makefile)
production-ready	CI workflow + stronger tests
research	Starter Jupyter notebook
enterprise	CI + pre-commit + notebook

Custom template	Effect
minimal	Essentials only
comprehensive	CI, pre-commit, notebook
microservice	API middleware patterns for FastAPI

🎯 Stack presets (Web UI cards)

Preset	Framework	Tracking	Orchestration	Deploy	Monitor
Quick Start	Sklearn	None	None	FastAPI	None
Data Science	Sklearn	MLflow	None	FastAPI	Custom
Deep Learning	PyTorch	W&B	None	Docker	None
Production MLOps	PyTorch	MLflow	Airflow	Docker	Evidently
Enterprise	TensorFlow	MLflow	Kubeflow	Kubernetes	Evidently
Research	PyTorch	W&B	None	FastAPI	None

---

🌐 Web UI (Recommended)

The fastest path: open the app, pick a stack preset, configure options, preview files, generate, and download a ZIP.

Run locally

cd web_ui
npm install
npm run dev
# Syncs templates/ from repo root, then starts Next.js
# → http://localhost:3000

API routes

Method	Route	Description
GET	/api/options	Dropdown values for the form
GET	/api/preview?framework=...&task_type=...	File tree preview (defaults for missing fields)
POST	/api/preview	Preview with partial body
POST	/api/preview?strict=true	Preview with full validation
POST	/api/generate	Start async generation → { task_id }
GET	/api/status/[taskId]	Poll progress
GET	/api/download/[taskId]	Download ZIP

Web UI tech stack

Layer	Technology
Framework	Next.js 16 (App Router, Turbopack)
UI	Tailwind CSS v4, Syne + DM Sans + JetBrains Mono
Forms	react-hook-form + Zod
Generator	TypeScript (web_ui/src/lib/generator/)
Templates	Jinja2 sources in templates/ (synced on dev/build)
Archive	archiver (ZIP)

Deploy / self-host

cd web_ui
npm run build   # runs scripts/sync-templates.mjs then next build
npm start

Set the deploy root to web_ui. The generator runs in API routes — no separate Python server required.

---

📦 Python CLI

Install and use the classic Typer CLI (renders templates/ with Jinja2):

pip install mlops-project-generator==2.1.0
# or from source:
pip install -e .

Interactive

mlops-project-generator init

With preset

mlops-project-generator init --preset production-mlops --project-name my-ml-platform
mlops-project-generator list-presets

Non-interactive (CI/CD)

mlops-project-generator init \
  --framework pytorch \
  --task-type classification \
  --tracking mlflow \
  --orchestration airflow \
  --deployment docker \
  --monitoring evidently \
  --project-name enterprise-ml \
  --author-name "ML Team" \
  --description "Production ML pipeline"

Advanced CLI (v2.x)

mlops-project-generator clone <source> --name <target>
mlops-project-generator archive <project> --type zip
mlops-project-generator check_deps --update --security
mlops-project-generator profile --format json
mlops-project-generator migrate <old> <new-framework>
mlops-project-generator doctor --fix --deep

init flags

Flag	Values
--framework / -f	sklearn, pytorch, tensorflow
--task-type / -t	classification, regression, timeseries, nlp, computer-vision
--tracking / -r	mlflow, wandb, custom, none
--orchestration / -o	airflow, kubeflow, none
--deployment / -d	fastapi, docker, kubernetes
--monitoring / -m	evidently, custom, none

---

📁 Generated project structure

Typical output (varies by stack, preset, and template):

your-project/
├── src/
│   ├── train.py                 # or nlp/finetune.py for NLP
│   ├── inference.py
│   ├── data/                    # loaders (+ NLP/CV helpers)
│   ├── models/                  # task-specific model modules
│   └── nlp/                     # Hugging Face (NLP task)
├── configs/
│   ├── config.yaml
│   └── nlp.yaml                 # NLP only
├── dags/                        # Airflow (if selected)
│   └── ml_training_dag.py
├── pipelines/                   # Kubeflow (if selected)
├── k8s/                         # Kubernetes (if selected)
│   ├── deployment.yaml
│   ├── service.yaml
│   └── configmap.yaml
├── scripts/monitoring/          # Evidently / custom monitoring
├── cloud/                       # Optional cloud provider pack
│   └── <provider>/<service>/
├── Dockerfile                   # Docker / K8s deploy
├── docker-compose.yml           # Docker deploy
├── .github/workflows/ci.yml     # production-ready / enterprise preset
├── tests/
├── notebooks/
├── requirements.txt
├── Makefile
├── .env.example
├── generation_insights.json     # if analytics enabled
├── project_config.json
└── README.md

---

🛠️ Framework & task highlights

Scikit-learn

• RandomForest / regression / time-series models via Jinja templates
• DataLoader, FeatureEngineering, joblib persistence

PyTorch & TensorFlow

• Task-specific model templates (classification_model, regression_model, timeseries_model)
• Training loops with tracking hooks; FastAPI inference when selected

NLP (Hugging Face)

• Fine-tune with transformers + datasets + Trainer
• Config-driven (configs/nlp.yaml), demo data if CSV missing
• FastAPI inference when deployment is FastAPI

Computer vision

• Image dataset helper + OpenCV/Pillow in requirements

---

🚀 Deployment & monitoring (generated)

Selection	Generated artifacts
FastAPI	src/inference.py with /health and /predict
Docker	Dockerfile, .dockerignore, docker-compose.yml, scripts/run_docker.sh
Kubernetes	k8s/*.yaml, scripts/deploy_k8s.sh
Evidently	scripts/monitoring/drift_report.py
Custom monitoring	scripts/monitoring/metrics_collector.py

---

🏗️ Repository structure

MLOps-Project-Generator/
├── generator/                 # Python CLI (Typer, renderer, cloud deployer, …)
├── templates/                 # Jinja2 sources (shared by CLI + Web UI sync)
│   ├── common/               # README, requirements, config, snippets, tests
│   ├── sklearn/ | pytorch/ | tensorflow/
│   └── nlp/                  # Hugging Face NLP templates
├── web_ui/                    # Next.js application
│   ├── app/api/              # generate, status, download, options, preview
│   ├── src/
│   │   ├── components/       # MLOpsForm, marketing, form sections
│   │   └── lib/
│   │       ├── generator/    # TS generator + legacy Jinja loader
│   │       ├── server/       # validation, generation pipeline
│   │       └── tasks/        # in-memory task store
│   └── scripts/
│       └── sync-templates.mjs  # copies ../templates → web_ui/templates
├── tests/                     # Python CLI tests
├── pyproject.toml
└── README.md

---

🧪 Development

Web UI

cd web_ui
npm install
npm run dev
npm run build

Python CLI

pip install -e ".[dev]"
pytest tests/ -v
black generator/ tests/
flake8 generator/ tests/

---

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Update templates/ and/or web_ui/src/lib/generator/ as needed
4. Run cd web_ui && npm run build and pytest for CLI changes
5. Open a pull request

---

📄 License

MIT — see LICENSE.

---

🗺️ Roadmap

Completed

• Python CLI with interactive + non-interactive modes
• Scikit-learn, PyTorch, TensorFlow templates
• MLflow, W&B, Airflow, Kubeflow, Docker, K8s, Evidently scaffolds
• Multi-cloud deployment templates (AWS, GCP, Azure)
• Web UI — Next.js generator with ZIP download
• TypeScript generator — no Python runtime on the server
• Legacy Jinja integration — templates/ rendered in Web UI for core ML tasks
• Hugging Face NLP — fine-tune + inference stubs
• Archive preview API — GET/POST /api/preview
• Deployment & orchestration files — real DAGs, K8s manifests, Docker Compose
• Config preset + custom template profiles in generated ZIPs
• UI redesign — marketing sections, stack presets, mobile compact layout
• Template quality pass — timeseries naming, model imports, requirements fixes

Upcoming

• Shareable project config links
• Template marketplace
• Additional frameworks (XGBoost, LightGBM)
• Deeper NLP/CV model templates in Jinja (beyond stubs)

---

📞 Support

• GitHub Issues
• GitHub Discussions

---

⭐ If this project helps you, consider starring the repo!

Generated with ❤️ by MLOps Project Generator