Build, train, and deploy machine learning systems with AI
AI/ML Engineering Framework encodes senior ML engineer patterns into prompt sequences that generate production-quality Python ML code. From data preprocessing to model deployment, every prompt produces clean, well-documented code that follows MLOps best practices. Includes a dedicated LLM fine-tuning module for the new era of AI engineering.
Full Access Unlocked
All 48 prompts · All 6 modules
"The problem framing prompts are something I wish I had 5 years ago. They force the right questions b..."
Dr. Fatima A. · ML Engineer
Need expert implementation?
Hire an OrchestratorConnect with a certified Prompt.Doctor Orchestrator to deploy this framework for you.
No coding required. You will use ChatGPT or Claude as your AI tool. Follow these steps in order — do not skip ahead.
Purchase & download the framework
Click the buy button on this page. After checkout, go to the and hit Download .zip. Unzip it — you'll get a .md file (the full framework) and a .pdf (easy to read reference). Keep both open.
Open your project — new or existing
This dashboard is designed to integrate into any existing project or be built as a standalone app. If you already have a site in Airo (or Cursor, Bolt, etc.), open that project. If you're starting fresh, create a new project. The Orchestrator Prompt handles both cases — it scans what's already there and adds only what's missing.
Paste the Orchestrator Prompt into your builder's chat
Open the on this page. Copy the Orchestrator Prompt and paste it into your AI builder's chat. It will scaffold the full admin system — secure login, email marketing module, booking engine, and CMS — on top of your existing codebase. This takes 2–5 minutes.
Add your API keys as secrets
Critical — Novice UsersIn your builder, go to Settings → Secrets and add the keys your app needs. For this framework: STRIPE_SECRET_KEY (for booking payments — get it from your Stripe dashboard), ANTHROPIC_API_KEY (for AI-assisted content — get it from console.anthropic.com), and DATABASE_URL (your MySQL connection string). No key is needed for the admin login, CMS, or email modules — those run on your existing infrastructure.
Don't have a MySQL server?
You can purchase a shared hosting plan with cPanel and MySQL at host.esgwon.dev. Once your account is set up, follow the step-by-step guide to create your database and connect it to your AI website builder.
How to set up cPanel MySQL & connect to your AI website →Need help with Stripe?
Get your STRIPE_SECRET_KEY and STRIPE_PUBLISHABLE_KEY from your Stripe dashboard. The guide covers test keys, webhook setup, and going live.
Need help with Anthropic?
Get your ANTHROPIC_API_KEY from console.anthropic.com. The guide covers model selection, cost management, and troubleshooting.
Prompting Airo after setup — always name the file
When asking Airo to add tables, columns, or features to the admin dashboard, always include src/server/lib/admin-db.ts in your prompt. Without it, Airo may target the wrong database. Example: "Add a bookings table in the admin database (src/server/lib/admin-db.ts) — do not touch any other database connection in this project."
Run the framework prompts inside your live app
Your app is now running in the builder's preview panel. Open the on this page, copy each prompt one at a time, and paste it into your builder's chat. Replace every [BRACKET] with your real data before sending. Work through the stages in order — each stage output feeds the next.
Test end-to-end, then publish or hand off to your client
Walk through the admin as a real user: log in, create a booking, send a test email campaign, update a CMS image, and run the Safe-to-Publish gate. Once everything passes, click Publish in your builder. Because this is a white-label dashboard, your client accesses it at /admin on their own domain — no Prompt.Doctor branding, no third-party login required.
6 modules · 48 prompts · 7 workflow stages
Problem Framing System
Translate a business problem into an ML problem statement. Includes task type classification, success metric definition, baseline establishment, and dataset requirements.
6 promptsData Pipeline Architecture
Generate data preprocessing pipelines: ingestion, cleaning, feature engineering, train/val/test splits, and data versioning with DVC.
9 promptsModel Architecture Generator
Generate PyTorch model architectures for classification, regression, NLP, computer vision, and time series tasks. Includes custom layer implementations.
10 promptsTraining & Evaluation System
Training loops with gradient clipping, learning rate scheduling, early stopping, W&B logging, and comprehensive evaluation frameworks.
10 promptsMLOps & Deployment
FastAPI model serving, Docker containerization, model versioning with MLflow, and cloud deployment prompts (AWS SageMaker, GCP Vertex AI).
7 promptsLLM Fine-Tuning System
LoRA and QLoRA fine-tuning prompts for Llama, Mistral, and Phi models. Includes dataset preparation, PEFT configuration, and evaluation.
6 promptsFrame the following business problem as an ML problem: [DESCRIBE BUSINESS PROBLEM]. Deliver: (1) ML task type (classification/regression/clustering/etc.), (2) Target variable definition, (3) Input features (what data we need), (4) Success metrics (primary and secondary), (5) Baseline to beat (simple heuristic), (6) Data requirements (volume, quality, labeling needs), (7) Key risks and failure modes, (8) Recommended model family and why. Be specific — avoid vague recommendations.
Generate a production-quality PyTorch training loop for a [TASK TYPE] model. Include: DataLoader setup with proper num_workers and pin_memory, Training and validation loops, Gradient clipping (max_norm=1.0), Learning rate scheduler ([SCHEDULER TYPE]), Early stopping with patience=[N], Weights & Biases logging (loss, metrics, learning rate, gradient norms), Model checkpointing (save best by validation metric), Reproducibility (seed everything), Mixed precision training with torch.cuda.amp. Add type hints and docstrings.
Generate a complete LoRA fine-tuning setup for [MODEL: Llama-3/Mistral/Phi-3] on a [TASK] dataset. Include: Dataset preparation with proper tokenization and chat template, LoraConfig with r=[RANK], alpha=[ALPHA], target_modules for this architecture, TrainingArguments optimized for [GPU TYPE], SFTTrainer setup from trl, Evaluation on held-out set, Model merging and saving for inference. Dataset format: [DESCRIBE FORMAT]. Training goal: [DESCRIBE GOAL].
system
13 prompts
workflow
9 prompts
architecture
10 prompts
prompts
16 prompts
"The problem framing prompts are something I wish I had 5 years ago. They force the right questions before a single line of code is written."
Dr. Fatima A.
ML Engineer
"The LLM fine-tuning module is exceptional. I fine-tuned a domain-specific model in a weekend using the LoRA prompts."
Ben W.
Data Scientist
All 48 prompts across 6 modules are unlocked for your account.
Lifetime access