THE HANDS-ON AI PROGRAMMING PROJECT

A Step-by-Step Guide to Building, Evaluating, and Deploying Machine Learning, Generative AI, and Agentic Applications

De: Desire Faucher

Narrado por: Virtual Voice

The Hands-On AI Programming Project: A Step-by-Step Guide to Building, Evaluating, and Deploying Machine Learning, Generative AI, and Agentic Applications

You can build an impressive AI demo—but can you turn it into software that works reliably beyond the notebook?

If you are a Python developer, software engineer, data professional, or aspiring AI engineer, you may already know how to train a model, call an API, or test a prompt. The harder challenge is combining those pieces into a complete AI application with dependable data pipelines, APIs, databases, retrieval, security, testing, deployment, and monitoring.

Without a clear engineering workflow, AI projects often become fragile collections of scripts. Models cannot be reproduced, prompts return inconsistent outputs, retrieval systems surface weak evidence, agents call unsafe tools, and deployment problems appear too late. The result is software that looks promising in development but is difficult to trust, maintain, evaluate, or release.

WHAT THIS BOOK OFFERS?

This book guides you through building a complete AI support and knowledge platform from the ground up. Instead of presenting disconnected examples, it develops one practical application across the full software lifecycle.

Using Python, FastAPI, PostgreSQL, scikit-learn, MLflow, generative AI, retrieval-augmented generation, Docker, GitHub Actions, and controlled agent workflows, you will learn how to:

• build reproducible machine learning pipelines and serve predictions through a production API

• create provider-neutral generative AI features with versioned prompts and validated structured outputs

• design a grounded RAG pipeline with document ingestion, hybrid search, reranking, permissions, and evidence checks

• implement safe AI agents with tools, approvals, budgets, idempotency, MCP integration, and recovery controls

• evaluate model quality, retrieval relevance, agent behaviour, security, latency, and cost before release

The project follows the practical BUILD workflow: bound the problem, understand the data, implement the smallest dependable solution, learn through evaluation, and deploy and diagnose. Each chapter extends the same codebase, includes verification checkpoints, and shows how individual AI components fit into a complete system.

This book is written for readers who want more than isolated tutorials and conceptual explanations. It is especially useful for developers moving into AI engineering, machine learning practitioners learning production software delivery, and backend engineers building LLM, RAG, or agentic applications.

The book’s value comes from its complete implementation structure. You will work through environment setup, architecture, data validation, model training, API development, structured generation, vector and keyword retrieval, tool execution, adversarial testing, release gates, containerisation, CI/CD, observability, feedback collection, model releases, multimodal workflows, and local model adapters.

The platform is designed around production concerns that are often missing from beginner AI projects: immutable model artefacts, transaction boundaries, tenant isolation, prompt injection defence, approval workflows, safe termination, tested rollback, distributed tracing, and measurable release thresholds.

By applying the methods in this book, you will gain the practical confidence to move from experimental AI code to maintainable, testable, and deployable AI software.

Build the complete project, understand every layer, and start creating AI applications that are engineered to work beyond the demo.