• Introduction to Generative AI
• What is Generative AI?
• AI vs ML vs Deep Learning vs Generative AI
• Predictive AI vs Generative AI
• Generative AI ecosystem
• Evolution of LLMs
• Foundation models
• Large Language Models
• Multimodal AI
• Generative AI use cases in business
• How LLMs Work
• Tokens
• Tokenization
• Embeddings
• Context windows
• Transformer architecture — conceptual understanding
• Attention mechanism
• Self-attention
• Positional encoding
• Parameters
• Pre-training
• Fine-tuning
• Instruction tuning
• RLHF
• Alignment
• LLM Capabilities and Limitations
• Hallucinations
• Knowledge limitations
• Context limitations
• Bias
• Reasoning limitations
• Prompt sensitivity
• Non-deterministic outputs
• Model selection considerations
• Practical
• Explore different LLMs
• Tokenization exercises
• Compare responses from different models
• Build a simple LLM-powered application

• Working with LLM APIs
• API architecture
• API keys and authentication
• Requests/responses
• System/user/assistant messages
• Chat completion concepts
• Structured responses
• JSON output
• Streaming responses
• Model Parameters
• Temperature
• Top-p
• Maximum tokens
• Stop sequences
• Seed
• Context window
• Model selection
• Working with Modern LLM Providers
• Students should get exposure to multiple model ecosystems, for example:
• OpenAI
• Google Gemini
• Anthropic Claude
• Open-source models
• Hugging Face ecosystem
• Multimodal GenAI
• Text → Text
• Text → Image
• Image → Text
• Audio → Text
• Text → Audio
• Video understanding
• Multimodal prompting
• Practical Project : Build a Multi-Model AI Chat Application

• Prompt Fundamentals
• Prompt anatomy
• Instructions
• Context
• Input
• Output requirements
• Constraints
• Role prompting
• Prompting Techniques
• Zero-shot prompting
• One-shot prompting
• Few-shot prompting
• Role prompting
• Contextual prompting
• Instruction prompting
• Delimiter techniques
• Advanced Prompting
• Chain-of-thought concepts
• Reasoning-oriented prompting
• Self-consistency
• Decomposition
• ReAct prompting
• Tree-of-thought concepts
• Reflection
• Critique and refinement
• Self-correction
• Structured Output
• JSON generation
• Schema-based output
• Function/tool calling
• Output validation
• Pydantic-based structured responses
• Prompt Optimization
• Prompt testing
• Prompt versioning
• Prompt templates
• Prompt evaluation
• Prompt injection awareness
• Practical Projects
• AI Resume Analyzer
• AI Content Generator
• Structured Data Extraction System

• Embeddings
• What are embeddings?
• Text embeddings
• Semantic representation
• Embedding dimensions
• Similarity
• Cosine similarity
• Semantic search
• Vector Databases
• Why vector dataases?
• Vector indexing
• Similarity search
• Metadata filtering
• Hybrid search
• Vector Database Ecosystem
• Hands-on exposure to:
• FAISS
• Chroma
• Pinecone
• Weaviate
• Qdrant
• Retrieval Strategies
• Top-K retrieval
• Similarity threshold
• Metadata filtering
• Dense retrieval
• Sparse retrieval
• Hybrid retrieval
• Practical Project : Semantic Search Engine (Students upload documents and search them using natural language.)

• RAG Fundamentals
• Why RAG?
• LLM limitations
• RAG architecture
• Retrieval vs generation
• RAG pipeline
• Document Processing
• PDF ingestion
• Word documents
• Excel
• CSV
• Web pages
• Text files
• Markdown
• HTML
• Document Chunking
• Why chunking?
• Fixed-size chunking
• Recursive chunking
• Sentence-based chunking
• Semantic chunking
• Parent-child chunking
• Chunk overlap
• Chunk-size optimization
• Embedding Pipeline
• Document → chunks
• Chunks → embeddings
• Embeddings → vector database
• Retrieval
• Similarity search
• Top-K retrieval
  Metadata filtering
• Context selection
• Context compression
• Advanced RAG
• Naive RAG
• Advanced RAG
• Modular RAG
• Hybrid RAG
• Query expansion
• Query rewriting
• Multi-query retrieval
• HyDE
• Reranking
• Reciprocal Rank Fusion
• Contextual retrieval
• RAG Evaluation
• Retrieval quality
• Context relevance
• Answer relevance
• Faithfulness
• Groundedness
• Precision
• Recall
• Practical Projects
• Project 1 — PDF Question Answering System
• Project 2 — Company Knowledge Base
• Project 3 — Website RAG Chatbot
• Project 4 — Enterprise Document Intelligence System

• LangChain Fundamentals
• LangChain architecture
• Models
• Prompts
• Output parsers
• Chains
• Retrievers
• Tools
• Memory
• Agents
• LangChain Expression Language
• Runnable concepts
• Runnable sequences
• Runnable parallel
• Runnable passthrough
• Pipeline construction
• LangGraph Introduction
• Why LangGraph?
• Stateful applications
• Nodes
• Edges
• State
• Conditional routing
• Graph-based workflows
• Other Framework Exposure
• Students should understand the ecosystem around: LlamaIndex, LangChain, LangGraph, Hugging Face

• Tools
• What is an AI tool?
• Tool definitions
• Tool selection
• Tool execution
• Tool responses
• Function Calling
• Function schemas
• Structured tool calls
• Tool execution loops
• Error handling
• Tool validation
• Building Tools
• Students build tools for: Calculator, Database, Web search, Weather, File search, Email, Calendar, REST APIs
• Tool Design
• Tool descriptions
• Tool boundaries
• Tool permissions
• Tool failures
• Tool security
• Project : AI Assistant with 5+ Real-World Tools

• Introduction to AI Agents
• What is an AI agent?
• LLM application vs AI agent
• Agent architecture
• Agent loop
• Perception
• Reasoning
• Planning
• Action
• Observation
• Memory
• Agent Patterns
• ReAct
• Tool-using agents
• Router agents
• Planner agents
• Reflection agents
• Critic agents
• Autonomous agents
• Human-in-the-loop agents
• Agent Memory
• Short-term memory
• Long-term memory
• Conversation memory
• Semantic memory
• Episodic memory
• Memory storage
• Agent Decision Making
• Tool selection
• Task decomposition
• Planning
• Replanning
• Error recovery
• Self-reflection
• Practical Projects
• AI Research Agent
• AI Travel Planning Agent
• AI SQL Agent
• AI Customer Support Agent

• Graph-Based Agent Architecture
• State
• Nodes
• Edges
• Conditional edges
• Checkpoints
• Persistence
• Multi-Step Agents
• Task decomposition
• Planning
• Execution
• Verification
• Retry
• Failure handling
• Multi-Agent Systems
• What is a multi-agent system?
• Supervisor architecture
• Worker agents
• Specialist agents
• Agent communication
• Agent delegation
• Agent collaboration
• Human-in-the-Loop
• Approval workflows
• Human review
• Interrupt/resume
• Sensitive action confirmation
• Projects
• Multi-Agent Research System
• Example: Research Agent → Data Analyst Agent → Fact Checker → Writer → Reviewer
• Another: Customer Query Agent → Product Agent → Billing Agent → Support Agent → Supervisor

• Topics :
• What is MCP?
• Why MCP?
• MCP architecture
• MCP clients
• MCP servers
• Tools
• Resources
• Prompts
• Context exchange
• MCP security
• Building MCP servers
• Connecting agents to MCP servers
• MCP with databases
• MCP with APIs
• MCP with enterprise systems
• Project
• Build an MCP Server + AI Agent
• Example: Agent → MCP → Company Database / Files / APIs

• Advanced RAG
• Query routing
• Query rewriting
• Multi-query RAG
• Hybrid search
• Reranking
• Metadata filtering
• Context compression
• Multi-vector retrieval
• Parent-document retrieval
• Knowledge graphs
• Agentic RAG
• Agent decides whether retrieval is required
• Agent chooses the data source
• Agent chooses retrieval strategy
• Agent evaluates retrieved information
• Agent performs additional searches
• Agent verifies answers
• Agent generates citations
• Advanced Architectures
•  

RAG Agent

User
↓
Agent
↓
Understand Query
↓
Choose Tool
├── Vector DB
├── Web Search
├── SQL DB
└── API
↓
Retrieve
↓
Evaluate
↓
Generate Answer
 

• Project : Enterprise Agentic RAG System

• Topics
• Text-to-SQL
• SQL agents
• Database tools
• Schema understanding
• Query generation
• Query validation
• Query execution
• SQL injection risks
• Data permissions
• Result interpretation
  Project : AI Data Analyst Agent ( Example: "Show me the sales performance of each region during the last six months." )
• Agent:
• Understands question
• Checks database schema
• Generates SQL
• Validates SQL
• Executes SQL
• Analyzes result
• Creates visualization
• Explains result

• Vision
• Vision-language models
• Image understanding
• OCR
• Document understanding
• Image-based RAG
• Audio
• Speech-to-text
• Text-to-speech
• Voice assistants
• Real-time conversations
• Video
• Video understanding
• Video summarization
• Video question answering
• Project : Multimodal Document Agent
  • Input: PDF + images + tables + scanned documents
    • Output: Natural-language answers with references.

• Evaluation
• LLM evaluation
• RAG evaluation
• Agent evaluation
• Golden datasets
• Test cases
• Regression testing
• Metrics
• Accuracy
• Faithfulness
• Context relevance
• Answer relevance
• Retrieval precision
• Retrieval recall
• Latency
• Cost
• Observability
• Prompt tracing
• Tool tracing
• Agent tracing
• Token monitoring
• Error monitoring
• Agent trajectory analysis
• Tools
• Exposure to: LangSmith, Langfuse, RAG evaluation frameworks
• Project : Build an Evaluation Pipeline for a RAG/Agent System

• Prompt injection
• Indirect prompt injection
• Jailbreaking
• Data leakage
• Sensitive information exposure
• Insecure tool usage
• Excessive agency
• Tool permission controls
• Authentication
• Authorization
• Input validation
• Output validation
• Data privacy
• Secure RAG
• Secure agents
• Practical : Students attack their own AI application and then implement safeguards.

• Production Architecture
• GenAI application architecture
• API architecture
• Model serving
• Vector database
• Caching
• Queues
• Databases
• Monitoring
• Optimization
• Token optimization
• Prompt optimization
• Model selection
• Latency optimization
• Cost optimization
• Caching
• Batching
• Smaller models
• Deployment
• FastAPI
• Docker
• Cloud deployment
• Environment variables
• Secrets management
• Logging
• Production RAG
• Indexing pipelines
• Incremental document updates
• Metadata management
• Access control
• Multi-user RAG

• Open-source vs closed models
• Hugging Face ecosystem
• Model selection
• Quantization
• GGUF
• LoRA concepts
• PEFT concepts
• Local LLMs
• GPU requirements
• Models to expose students to
• Depending on the current ecosystem:
• Llama-family models
• Qwen-family models
• Mistral-family models
• Gemma-family models
• Local AI
• Ollama
• Local inference
• Model serving
• Project : Build a Private Local RAG Assistant

• Prompt engineering vs RAG vs fine-tuning
• When to fine-tune
• Supervised fine-tuning
• Instruction datasets
• Dataset preparation
• LoRA
• QLoRA
• PEFT
• Fine-tuning workflow
• Evaluation
• Model deployment
• Mini Project : Fine-tune an Open-Source LLM for a Domain-Specific Task

Students should complete at least 2 major projects.

CAPSTONE 1 — Enterprise Knowledge Agent
Documents
  ↓
Document Processing
  ↓
Chunking
  ↓
Embeddings
  ↓
Vector DB
  ↓
RAG
  ↓
Agent
  ↓
Tools
  ↓
Answer + Citations

Features:

PDF/Word ingestion
RAG
Hybrid search
Reranking
Agentic retrieval
Tool calling
Conversation memory
Authentication
Evaluation
Monitoring
CAPSTONE 2 — Multi-Agent Business Automation System

Example:

AI Business Operations Agent

User
↓
Supervisor Agent
├── Research Agent
├── Data Analyst Agent
├── Database Agent
├── Email Agent
├── Report Agent
└── Reviewer Agent
       ↓
    Final Result

Features:

• Multi-agent architecture
• Planning
• Tool calling
• RAG
• MCP
• Memory
• Human approval
• Evaluation
• Security
• Deployment
   

------------------------------------------

Recommended Technology Stack :

• Area    Technologies
  LLM APIs   :  OpenAI, Gemini, Claude
• Open-source AI    : Hugging Face
• Local LLM   :  Ollama
• GenAI Framework  :  LangChain
• Agent Framework  :  LangGraph
• RAG   :  LangChain / LlamaIndex
• Vector DB   :  Chroma / FAISS / Qdrant / Pinecone
• Embeddings  :  OpenAI / Hugging Face
• Reranking  :  Cross-encoder / modern rerankers
• MCP  :  Model Context Protocol
• API  :  FastAPI
• Database  :  PostgreSQL
• SQL Agent  :  LangChain/LangGraph
• Evaluation  :  RAGAS / LangSmith / Langfuse
• Deployment  : Docker + Cloud
• Version Control   : Git + GitHub