Subcribe and Access : 5200+ FREE Videos and 21+ Subjects Like CRT, SoftSkills, JAVA, Hadoop, Microsoft .NET, Testin5g Tools etc..

Batch Date: May 6th @7:30AM

Faculty: Mr. Vasanth (8+ Yrs of Exp,..)

Duration: 3 Months

Venue :
DURGA SOFTWARE SOLUTIONS,
Flat No : 202, 2nd Floor,
HUDA Maitrivanam,
Ameerpet, Hyderabad - 500038

Ph.No: +91 - 8885252627, 9246212143, 80 96 96 96 96

Syllabus:

AGENTIC AI Engineering
with GEN AI Foundations

Module 1: Generative AI & LLM Foundations

Objective: Understand how Large Language Models work, call LLM APIs, run open-source models locally, and build semantic search systems.

1.1 Introduction to Generative AI

• What is Generative AI? Generative vs Discriminative models
• AI landscape 2026: OpenAI, Anthropic, Google DeepMind, Meta, Mistral
• Types of AI: Narrow AI, General AI, Superintelligent AI
• Why Generative AI matters — real-world applications and business value
• AI vs Machine Learning vs Deep Learning vs Generative AI

1.2 How LLMs Work

• Transformer architecture and attention mechanism — plain English explanation
• Tokens, context window, temperature, top-k, top-p sampling
• Tokenization and embeddings — foundational overview
• LLM families: GPT-4o, Claude, Gemini, LLaMA 3, Mistral, DeepSeek
• Open-source vs proprietary models — model selection framework

1.3 LLM APIs & Prompt Engineering

• Calling OpenAI and Anthropic APIs: system, user, assistant messages
• Multi-turn conversation state management in Python
• Prompt Engineering techniques:
  • Zero-shot, few-shot, and chain-of-thought (CoT) prompting
  • ReAct prompting: Reason + Act pattern
  • Tree-of-Thought (ToT), persona-based, constraint-based prompting
  • Meta-prompting, negative prompting
• Context Engineering: system prompt design, goal framing, constraint embedding
• Evaluating LLM outputs — quality metrics and best practices
  

1.4 Open-Source LLMs & Local Deployment

• Open-source LLM families: LLaMA 3, Mistral, DeepSeek, Phi
• Running models locally with Ollama — privacy and cost benefits
• Comparing open-source and proprietary models for different use cases

1.5 Embeddings & Vector Databases

• What are embeddings? Semantic similarity and cosine distance
• Embedding models: OpenAI text-embedding-3, sentence-transformers
• Vector databases:
  • ChromaDB — local vector store for development
  • FAISS — Facebook AI Similarity Search
  • Pinecone — managed vector database for production
• Indexing, similarity search, and metadata filtering
• Hands-on project: semantic document search engine
• Hands-on Project: Multi-turn chatbot using LLM API + Semantic document search engine

Module 2: LLM Automation, Chains & Retrieval-Augmented Generation (RAG)

Objective: Build end-to-end LLM automation pipelines and production-grade RAG systems with full evaluation.

2.1 LangChain Architecture & LCEL

• LangChain overview: LLMs, Chains, Prompts, Memory, Agents, Tools
• LangChain Expression Language (LCEL): the modern pipeline syntax
• Chain types:
  • LLMChain, SimpleSequentialChain, SequentialChain
  • ConversationChain, RouterChain
• Conditional branching and dynamic routing pipelines
• Prompt Templates: Standard, Few-shot, Zero-shot, and Custom templates
• Document Loaders and Text Splitters: preparing external knowledge for LLMs

2.2 Agent Cognitive Layers & Memory

• Agent cognitive architecture: Perception → Memory → Decision → Action
• Pydantic data contracts between cognitive modules
• Memory types:
  • ConversationBufferMemory
  • ConversationSummaryMemory
  • ConversationBufferWindowMemory
  • VectorStoreRetrieverMemory
• Output parsers: Pydantic, JSON mode, structured schema-enforced outputs
• Python logging for agents: DEBUG, INFO, WARNING, ERROR, CRITICAL levels

2.3 Retrieval-Augmented Generation (RAG)

• What is RAG and why it is needed — the hallucination problem
• RAG architecture: Naive RAG → Advanced RAG → Agentic RAG
• Components of RAG:
  • Document ingestion: PDF, web, CSV, Notion loaders
  • Chunking strategies: fixed-size, semantic, recursive, parent-child
  • Vector Database (FAISS, Pinecone, ChromaDB)
  • LLM integration for answering queries
• Retrieval techniques:
  • Maximum Marginal Relevance (MMR)
  • Multi-query retriever
  • Contextual compression
  • HyDE — Hypothetical Document Embeddings
• ReRanking with Cohere cross-encoders
• Hybrid Planning: Heuristics + LLM decision matrix for reliable retrieval
• Function calling and Tool use: OpenAI tools schema, Anthropic tool_use
• Applications of RAG: chatbots, search, knowledge assistants, enterprise Q&A;
• Workflow of RAG — step-by-step explanation

2.4 RAG Evaluation & Observability

• RAGAS evaluation framework:
  • Answer Relevance
  • Context Precision
  • Faithfulness
  • Answer Correctness
• LangSmith: end-to-end pipeline tracing, cost tracking, latency profiling
• Evaluation-driven iteration: diagnosing and fixing failing RAG pipelines
• Hands-on Project: Full RAG chatbot with source citations, RAGAS evaluation scores, and LangSmith observability

Module 2.5: Knowledge Graphs & GraphRAG (Neo4j)

Objective: Understand enterprise knowledge graph design with Neo4j and build GraphRAG systems that combine vector search with graph traversal for complex multi-hop reasoning.

2.5.1 Knowledge Graph Fundamentals (Neo4j)

• Why knowledge graphs? When vector search alone fails
• Entity-relation modeling: nodes, edges, properties
• Ontology-driven design: defining schema for AI systems
• Provenance and audit trails in graphs
• Schema design and constraints for enterprise AI
• Neo4j: Cypher query language basics
• Structured reasoning using relationships
• Explainability via graph traces — show the reasoning path
• Governance-ready data representations

2.5.2 GraphRAG — Hybrid Retrieval with Neo4j

• GraphRAG vs traditional RAG: when to use which
• Graph traversal retrieval: multi-hop reasoning
  • Example: "Who approved this contract, what team do they manage, and what policy governs it?"
  • Vector search cannot answer this — graph traversal can
• Hybrid retrieval: vector + full-text + graph traversal combined
• "Show evidence + show path" response pattern — explainable answers
• Relationship-aware retrieval for complex enterprise questions
• Ingesting documents → extracting entities → building Neo4j provenance graph
• Evaluation: groundedness, citation coverage, retrieval recall
• Hands-on Project: GraphRAG agent — ingest documents, extract entities to Neo4j, hybrid retrieval with explainable evidence + graph path

Module 3: Agentic AI Foundations & Design Patterns

Objective: Understand the core building blocks and architectural patterns of autonomous AI agents and build single-agent systems using multiple frameworks.

3.1 Introduction to Agentic AI

• What is Agentic AI? Agent vs Chain — the key distinction
• Core AI agent building blocks: Perception, Cognition/Reasoning, Planning, Action, Memory, Adaptability & Learning
• Agent anatomy: Perception → Reasoning → Action loop
• LLM as the brain, tools as hands, memory as context
• Agent vs traditional software: autonomous decision-making explained

3.2 Agentic AI Architectures & Design Patterns

• Architectural concepts: single-agent, multi-agent, hierarchical, swarm
• Key Design Patterns:
  • ReAct — Reason + Act pattern: Thought → Action →Observation loop
  • Reflection — agents that critique and self-improve their own outputs
  • Tool Use — structured tool calling, error handling, max-iteration safety
  • Planning — Plan-and-Execute, Tree-of-Thought, ReWOO
• Human-in-the-Loop (HITL): when and how to add human oversight
• Hybrid planning: combining rule-based heuristics with LLM reasoning

3.3 Foundational Frameworks & Technologies

LangChain and LangGraph

• ReAct agents with LangChain AgentExecutor
• Tool selection logic and real-world tools: Tavily web search, calculator, database
• LangGraph: stateful workflows with graph nodes, edges, and TypedDict state
• DAG execution principles: topological sort, fallback nodes, parallel variants
• Human-in-the-loop: interrupt_before, checkpointers (SQLite, Postgres), workflow resumption

OpenAI Agents SDK

• Agents, Handoffs, Guardrails, Tracing
• Multi-step agents with tool registration and streaming
• Handoff patterns: routing between specialised agents
• Comparison: LangChain AgentExecutor vs OpenAI Agents SDK

3.4 Practical Agent Development & Deployment

• Building agents with Python
• Tools and APIs integration
• Context Engineering for agents: system prompt design and goal framing
• Agent reliability: max iterations, fallback strategies, safe termination
• Hands-on Projects: Research Agent (ReAct + OpenAI SDK) and Email-Draft Agent with human approval gate (LangGraph)
  

Module 4: Multi-Agent Systems, A2A Protocol & Frameworks

Objective: Build production-grade multi-agent systems using CrewAI, AutoGen, and n8n. Implement A2A protocol for cross-framework agent interoperability.

4.1 Multi-Agent Systems

• Why multiple agents? Specialisation, parallelism, and fault isolation
• Communication patterns: broadcast, blackboard, supervisor, swarm
• Orchestrator vs subagent roles and responsibilities
• Inter-agent memory sharing and context passing

CrewAI & Multi-Agent Systems

• CrewAI architecture: Agents, Tasks, Tools, Crews
• Process types: sequential, hierarchical, parallel
• Designing agent roles, goals, and backstories
• CrewAI Flows: event-driven orchestration and multi-crew state management
• Hands-on projects:
  • 3-agent Content Pipeline: Researcher + Writer + Editor
  • Stock Picker Agent: research, analyse, and recommend investments
    

4.2 AutoGen — Conversational Multi-Agent & Code Agents

• AutoGen architecture: ConversableAgent, AssistantAgent, UserProxy
• GroupChat and GroupChatManager for multi-agent conversations
• Code execution agents: write, run, and debug code autonomously in Docker sandbox
• Agent Factory Pattern: reusable, scalable agent blueprints for any domain
• Specialised agents:
  • Browser Automation Agent using Playwright
  • Database Agent for data querying and analysis
  • API Testing Agent
• Hands-on project: 4-agent Engineering Team — PM + Developer + Tester + Reviewer

4.3 A2A Protocol — Agent-to-Agent Communication

• What is the A2A Protocol? Google → Linux Foundation open standard
• A2A vs MCP: MCP = agent-to-tool communication, A2A = agent-to-agent communication
• A2A architecture:
  • Agent Cards: capability discovery via JSON at well-known URLs
  • Client-server model: A2A client (delegating agent) and A2A server
    (remote agent)
  • Task lifecycle: states, progress tracking, result artifacts
  • Authentication and secure inter-agent communication
• A2A ecosystem: 150+ enterprise partners including Salesforce, SAP, ServiceNow, LangChain
• Building A2A-compliant agents with CrewAI, LangGraph, and Google ADK
• Hands-on: two agents on different frameworks communicating via A2A protocol

4.4 No-Code/Low-Code Agent Development

n8n — Visual Agentic Workflow Builder

• What is n8n? Visual workflow builder with AI nodes and triggers
• Integrations: Gmail, Google Sheets, Slack, Google Calendar, webhooks
• AI nodes: LLM calls, classification, summarisation, response generation
• No-code agent development: build workflows without writing Python
• Hands-on: email triage workflow — classify → respond → log in spreadsheet
  Cloud deployment of n8n workflows

Module 5 : Responsible AI & Evaluation

Objective: Understand ethical principles and safety frameworks for AI systems. Implement observability and evaluation pipelines. Apply technical guardrails in production agents.

5.1 Responsible AI and Evaluation

• Observability and evaluation:
  • RAGAS deep evaluation: Answer Relevance, Context Precision, Faithfulness, Answer Correctness
  • LangSmith end-to-end agent tracing and cost tracking
  • AgentOps: cost dashboard, token budget tracking, latency profiling
  • Prompt versioning and registry management
  • Evaluation-driven iteration: diagnosing and fixing agent failures
• Ethical considerations and risk mitigation:
  • Bias and fairness in LLM outputs: detection and mitigation
  • Transparency, explainability, and accountability frameworks
  • EU AI Act: risk tiers awareness
    (prohibited, high-risk, limited-risk, minimal-risk)
  • Who is responsible when an agent fails? Accountability in enterprise AI

5.2 Security Threat Modeling for AI Agents

• Threat model for production agents: the 4 main attack surfaces
  • Prompt injection: malicious instructions injected via user input or retrieved documents
  • Indirect prompt injection: attacker-controlled content in the agent's context window
  • Unsafe tool calls: agent tricked into executing destructive or unintended actions
  • Data exfiltration: agent leaking sensitive retrieved content to untrusted endpoints
• Architectural defence patterns:
  • Input sanitisation and allowlist-based tool access (read-only by default)
  • Views-only execution patterns: never query base tables or execute write operations without approval
  • Human approval gates: approve / edit / reject before agent executes high-risk actions
  • Least-privilege tool design: agents get minimum permissions required
  • Auditable tool execution: every tool call logged with inputs, outputs, timestamps
• Prompt injection: definition, real attack examples, and step-by-step defence strategies
• Adversarial prompting and jailbreak attempts — patterns and mitigations
• Security testing: red-teaming your agents before production release

5.3 Technical Guardrails Implementation

• Guardrails AI: input and output validation, topic rails
• NeMo Guardrails: conversation flows, topical rails, safety rails
• PII (Personally Identifiable Information) detection and masking
• Hallucination detection and factual grounding
• OpenAI Agents SDK guardrails: built-in safety parameters
• From ethics to code: translating responsible AI principles into guardrail design

Module 6: Production Deployment & MLOps

Objective: Instrument AI agent systems for production, deploy to live endpoints, and understand the MLOps landscape.

6.1 Practical Agent Deployment

• Building agents with Python, Tools, and APIs
• FastAPI: wrapping agents as REST API endpoints
• Docker: containerising agent applications
• Environment management: .env files, secrets, API key security
• Cloud deployment:
  • HuggingFace Spaces — free, portfolio-ready deployment
  • Render / Railway — simple backend deployment
  • AWS EC2 + S3 — cloud deployment basics
• Streamlit and Gradio: building agent UIs for demos and production

6.2 Model Context Protocol (MCP)

• What is MCP? Anthropic's open standard for agent-to-tool communication
• MCP architecture: MCP server, MCP client, tool schema standardisation
• Building an MCP server hands-on: exposing tools via MCP
• MCP + A2A together: complete protocol stack for enterprise agents

6.3 Cost Optimisation & MLOps Awareness

• Semantic caching with GPTCache — reduce cost and latency
• LLM routing: strong model vs fast model based on query complexity
• CI/CD for AI: conceptual overview with a GitHub Actions example
• A/B testing agents: comparing prompt and model versions
• Drift detection: monitoring output quality over time
• What comes after this course: MLOps Engineering roadmap

Module 7: Capstone Projects & Future of Agentic AI

Objective: Build and present an end-to-end production-grade AI system. Understand where the field is heading and plan your career roadmap.

7.1 Capstone Projects

• Students choose one capstone project from five options:
  • Intelligent Document Assistant: multi-format RAG with Streamlit UI, citations, and ReAct agent
  • Autonomous Research Reporter: CrewAI multi-agent — Researcher + Analyst + Writer
  • AI Workflow Automation Bot: LangGraph agent with human approvals, MCP + A2A
  • 4-Agent Engineering Team: AutoGen crew — PM + Developer + Tester + Reviewer
  • Domain Expert Agent: RAG + tool agent in HR, finance, legal, or healthcare with guardrails
• Deliverables: working demo at live URL, RAGAS scores, architecture doc, GitHub repo, 5-min presentation

7.2 Future of Agentic AI & AGI

• Where Agentic AI is heading: reactive ® proactive ® autonomous organisations
• The evolving agent protocol stack:
  • MCP: agent-to-tool communication standard (Anthropic, 2024)
  • A2A: agent-to-agent communication standard
    (Google/Linux Foundation, 2025)
  • What comes next: emerging protocols and standards
• AGI timeline: honest framing, current state, and what it means for practitioners
• Career pathways in Agentic AI Engineering:
  • AI Engineer
  • MLOps / AI DevOps Engineer
  • AI Product Manager
  • AI Solutions Architect
• Skills that will remain relevant vs what will be automated away
• Course 2 roadmap: Agentic AI in Production & MLOps Engineering