Building Agentic AI: Workflows, Fine-Tuning, Optimization, and Deployment by Sinan Ozdemir

Building Agentic AI: Workflows, Fine-Tuning, Optimization, and Deployment by Sinan Ozdemir

The book is intentionally written for "builders"—developers, data scientists, and technical product managers who need to design systems that handle real-world tasks without human hand-holding. The text is systematically divided into three tactical acts:Defining what separates a basic text-completion prompt or chatbot from a true "agent" (the capability to respond, reason, plan, and execute actions independently)Building foundational, deterministic LLM-driven sequences—such as complex text-to-SQL data pipelines—before layering on autonomous decision-making.Moving beyond vibes-based prompt engineering. Ozdemir introduces hard frameworks to measure precision, recall, and latency, demonstrating how to systematically benchmark models.

Ozdemir’s text arrives precisely as the software industry transitions out of the generative AI hype phase and into the hard reality of deploying functional production pipelines.The book refuses to pretend that building enterprise AI is easy. Ozdemir pulls back the curtain on hidden architectural flaws that plague complex setups, such as positional bias in tool selection (where an LLM ignores an option simply based on where it sits in a prompt list, sometimes shifting accuracy by 40%). Rather than ignoring these engineering edge-cases, he provides clear algorithmic mitigation techniques.

One of the costliest mistakes engineering teams make is over-engineering simple problems into unpredictable agentic loops. Ozdemir provides clear architectural rubrics determining when a task requires a flexible, self-directing autonomous agent and when it should be constrained into a predictable, structured workflow.

Running unoptimized, massive foundation models across thousands of enterprise workflows will bankrupt a standard infrastructure budget. Building Agentic AI is highly celebrated for its focus on model distillation and optimization. By teaching developers how to combine smaller, fine-tuned models, speculative decoding, and highly adaptive vector spaces, Ozdemir shows how to match frontier-model performance at a fraction of the token cost.

Language: English.

Genre: Applied Artificial Intelligence.

Binding: সেলাই করা বাইন্ডিং

Quality: Premium Quality Books.

Printing: High Quality Printing.

Paper: Eye Friendly paper (Cream White)

Cover: Matt cover (Paperback).