Course Objective & Positioning

This syllabus provides a deep, implementation‑focused curriculum for building production‑grade Agentic AI systems using Python, Large Language Models (LLMs), and Retrieval Augmented Generation (RAG).

The course is designed for engineers who want to move beyond chatbots and build autonomous, reliable, governed AI systems used in real enterprises.

The focus is on system architecture, coding practices, reliability, safety, and hands‑on implementation.

Target Audience

Python developers (3–12+ years experience)

Backend, data, and platform engineers transitioning to AI roles

Senior engineers aiming for Agentic AI Lead / Architect positions

Module 1: Python Engineering Foundations for Agentic AI

Advanced Python patterns for long‑running, stateful systems

Async programming (async / await) for agent workflows

Concurrency, GIL considerations, and I/O bound workloads

Configuration management and secret handling

Structured logging, metrics, and distributed tracing

Writing testable, maintainable, production‑grade Python services

Lab: Build a resilient Python service with retries, logging, and async execution

Module 2: Large Language Models (LLMs) – Engineering View

How LLMs work internally (tokens, attention, context windows)

Prompting vs system‑driven design

Temperature, top‑p, deterministic vs non‑deterministic outputs

Function and tool calling

Structured JSON‑first responses for automation

Latency, throughput, and cost trade‑offs

Lab: Build a Python LLM wrapper with structured outputs and error handling

Module 3: Retrieval Augmented Generation (RAG)

Why hallucination happens and how RAG solves it

Embedding lifecycle and vector representations

Chunking strategies and relevance optimization

Vector databases (FAISS, Pinecone, Azure AI Search)

Hybrid search (vector + keyword)

Metadata filtering and access control

RAG evaluation metrics and feedback loops

Lab: Build a full RAG pipeline with vector search and evaluation

Module 4: Agentic AI Architecture & Reasoning Core

What makes an AI system truly agentic

Goal‑driven vs prompt‑driven systems

Agent core loop: Goal → Plan → Act → Observe → Learn

Planner vs executor separation

Task decomposition and hierarchical planning

Dynamic replanning on partial failure

Stop conditions and infinite‑loop prevention

Failure handling and recovery strategies

Lab: Implement a planner‑first agent with failure recovery

Module 5: Tool‑Using Agents & Memory Systems

Designing tool‑capable agents (APIs, databases, jobs)

Read‑only vs write tools and risk classification

Runtime tool selection and safe execution

Idempotency, retries, and partial execution handling

Short‑term, long‑term, and episodic memory

Memory summarization and decay strategies

Learning from past outcomes

Lab: Build a tool‑using agent with episodic memory and approvals

Module 6: Multi‑Agent Systems & Coordination

Why single agents fail at scale

Supervisor‑worker architecture

Domain‑specialized agents

Peer collaboration and debate patterns

Conflict detection and arbitration

Voting, confidence scoring, and escalation

Deadlock and livelock prevention

Lab: Implement a multi‑agent system with conflict resolution

Module 7: Enterprise Safety, Governance & AgentOps

Why governance is mandatory for agentic systems

Human‑in‑the‑loop approval models

Role‑based access control for agents

Tool permissioning and scoping

Budget and cost enforcement

Action‑level observability and audit trails

Behavior drift detection and alerts

Deterministic replay and compliance readiness

Lab: Build an approval workflow and agent observability dashboard

Capstone Project: Retail Agentic AI System (End‑to‑End)

Business problem: demand volatility, stock‑outs, and overstocking

Multi‑agent design: sales agent, inventory agent, demand forecast agent, supervisor agent

RAG over historical sales, supplier data, and policy documents

Planner‑driven decision making

Human approval for high‑risk actions

Memory‑based learning from outcomes

End‑to‑end demo and evaluation

Final Deliverable: production‑ready Agentic AI system

Learning Outcomes & Career Impact

Ability to design and build production‑grade agentic AI systems

Strong understanding of autonomy, reliability, and governance

Hands‑on experience with real enterprise patterns

Preparation for Senior Agentic AI Engineer, AI Pl