AGENTIC AI
WITH PYTHON
An LLM is just the start. Go from a simple API call to a working agent. Add structured output, context, memory, RAG, state, tools and LangGraph workflows to your own Python software.
See what you'll build ↓Not just an LLM behind a chat box. A real agent in your software.
An agentic application combines LLM, context, memory, knowledge, state, tools and workflow. You'll build each piece, then wire them together.
What you'll walk away with
Practical patterns you can apply to your existing software solutions.
Seats are limited for hands-on mentoring. Join the priority list and you'll hear the date first, before it goes public.
GET THE DATE FIRST→Built for professionals who put AI into real software
What will you actually build?
Four practical layers take you from a single LLM call to a full agent workflow.
Structured Output
Make an LLM return validated, predictable data your application can trust, such as a project status with open bugs.
Memory Assistant
Extract user preferences, store them persistently in SQLite and use them in later conversations.
RAG Assistant
Build an HR policy assistant that retrieves from documents and answers with grounded, source-aware responses.
LangGraph Agent
Combine LLM, context, memory, RAG, state, tools and tool selection into one working agent workflow.
16 modules of code-first engineering
Each concept is built step by step in Python and Jupyter Notebook, then combined into one integrated agent architecture.
- What is Agentic AI?
- From conventional software to AI-enabled software
- Role of LLMs in software solutions
- Why LLMs alone are not sufficient
- Introduction to agent-based software solutions
- LLM integration
- Sending requests
- Receiving responses
- Using responses in application logic
- Basic LLM interaction
- Context
- Memory
- Retrieval
- State
- Tools
- Agent workflow
- LLM integration
- Context
- Memory
- RAG
- State
- Tools
- Design patterns
- Agent architecture
- Capstone
- Python classes and objects
- Functions
- Dictionaries and lists
- JSON
- External libraries
- Python files
- Jupyter notebooks
- Interactive execution
- Code cells
- Markdown cells
- Incremental execution
- Inspecting results
- Required libraries
- LLM client
- API credentials
- Running notebooks
- LLM client
- Model
- Requests
- Responses
- System instructions
- User requests
- Separating instructions and input
- User request → Application → LLM → Response
- Text
- JSON
- Structured responses
- Application use
- Free-form response problems
- Predictable application data
- Models
- Fields
- Instances
- Validation
- JSON input
- Conversion
- Validation errors
- Generating schemas
- Expected output
- Strict structured output
- LLM → JSON Schema → Structured Response → Pydantic Model
- Structured project status: project name, status, completion percentage, open bugs
- What is context?
- Why agents require context
- Context versus user request
- System instructions
- Conversation history
- Tool results
- Retrieved information
- Task state
- User request
- Combining information from multiple sources
- Customer support scenario using conversation, transaction information, policy, state, and LLM
- Context size
- Large-context problems
- Relevant information retrieval
- Build a context-aware customer support assistant
- Why agents need memory
- Future-useful information
- Conversation versus persistent information
- LLM-based preference extraction
- Preference key and value
- Memory candidates
- Python representation
- Session memory
- Context integration
- Stored preferences → Relevant memory → Context → LLM → Response
- SQLite
- Memory table
- Storage
- Retrieval
- Persistent retrieval
- Formatting memory
- Providing memory to LLM
- Build an assistant that stores and later uses user preferences
- External information
- Retrieval before generation
- RAG architecture
- Source
- Page
- Content
- Searching
- Matching
- Selecting relevant content
- Question → Retrieval → Information → LLM → Answer
- Use retrieved information
- Handle unavailable information
- Embeddings
- Vector databases
- Semantic retrieval
- ChromaDB
- Persistent client
- Collections
- Documents
- Metadata
- Queries
- Results
- Source
- Page
- Build an HR policy assistant
- Structured information describing current agent execution
- Maintaining information
- Tracking known information
- Missing information
- Current task
- Multi-step workflows
- Pydantic state model
- Goal
- Brand
- Screen size
- RAM
- Required
- Optional
- Missing
- Existing
- Determine whether required information is complete
- Receive message
- Examine state
- Extract information
- Update state
- Preserve existing information
- Incremental information
- Ask only for missing information
- Build a multi-turn shopping requirements assistant
- Agent tools
- Why LLMs need tools
- Reasoning versus execution
- Project status
- Project bugs
- Project teams
- Name
- Description
- Function
- Arguments
- Collection
- Listing
- Descriptions
- Catalog generation
- Lookup
- Identification
- Execution
- Tool name + arguments → registry → function → result
- Unknown tools
- Invalid names
- Invalid arguments
- Build a project-assistant tool registry
- Multiple tools
- Relevant tool
- Required arguments
- Tool name
- Arguments
- Project name
- Available tools
- Descriptions
- Arguments
- User request
- User request → LLM → Tool selection → Arguments
- Pass selection to registry
- Implement select_tool()
- Request → Selection → Tool → Arguments → Execution → Result
- Build a project assistant with LLM-based tool selection
- Combine context, memory, RAG, state, tools, and selection
- Select Tool
- Execute Tool
- Generate Answer
- Request
- Tool name
- Arguments
- Result
- Answer
- Select Tool Node
- Execute Tool Node
- Generate Answer Node
- START → Select Tool → Execute Tool → Generate Answer → END
- Implement a multi-node project assistant
- Graph-based workflows
- Nodes
- State
- Edges
- START
- END
- Create StateGraph
- Structured workflow state
- Python node functions
- Read and update state
- Connect workflow nodes
- Build and compile graph
- Provide request
- Execute workflow
- Read result
- Convert project assistant to LangGraph
- Why patterns matter
- Reusable agent patterns
- Separation of responsibilities
- LLM → Schema → Validated object
- Sources → Context → LLM
- Conversation → Memory → Storage → Retrieval → Agent
- Question → Retrieval → Information → LLM → Answer
- Request → Selection → Execution → Result
- Interaction → State update → Missing information → Next action
- State + Nodes + Edges → Agent workflow
- Combine multiple patterns into capable agents
- Major components of an Agentic AI solution
- User
- Application
- LLM
- Context
- Memory
- RAG
- State
- Tools
- Registry
- Workflow
- External systems
- Request processing
- Decision making
- Structured output
- Tool selection
- Instructions
- History
- Retrieved information
- Tool results
- Task state
- Short-term
- Long-term
- Persistent storage
- Retrieval
- Documents
- Retrieval
- Vector database
- Relevant information
- Definitions
- Catalog
- Registry
- Execution
- State
- Nodes
- Edges
- Execution
- Interaction with existing software through tools
- Combine course concepts
- LLM
- Structured output
- Context
- Memory
- RAG
- State
- Tools
- Selection
- Workflow
- Request → Understand → Read state/context → Retrieve → Select tool → Execute → Update state → Generate response
- Required state
- Available tools
- Knowledge
- Memory
- Nodes
- Transitions
- Design and implement an Agentic AI application demonstrating the course concepts
- LLM
- Structured output
- Context
- Memory
- RAG
- State
- Tools
- Tool selection
- Workflow
- User interaction
- Agent state
- Context
- Memory
- Knowledge/RAG
- Tools
- Workflow
- Define problem
- Define interaction
- Define state
- Integrate LLM
- Structured responses
- Context
- Memory
- RAG
- Tools
- Tool selection
- Execution
- Workflow
- Integration
- Testing
- User interaction
- Decision process
- State changes
- Tool selection
- Tool execution
- Retrieval
- Memory
- Final response
- Workflow
13 hands-on exercises. Skills you can apply immediately.
Practice by building, step by step
Every exercise builds on the last, ending in your capstone.
By the end you'll be able to
Practical skills you can apply immediately.
- Integrate an LLM into Python applications and validate its structured output
- Design context and implement short-term and persistent memory
- Implement RAG with vector retrieval using ChromaDB
- Manage agent state across multi-step, multi-turn workflows
- Build a tool catalog and registry with LLM-based tool selection
- Build agent workflows in LangGraph and apply Agentic AI design patterns
Taught by someone who ships this for a living
Thayanithy Jegan
Learn from a technologist who works on both the engineering side of AI and its practical application inside real organisations — not a slide deck written by marketing.
Invest in skills that pay off from day one
16 modules, 13 hands-on exercises and a capstone. HRD Corp claimable, certificate included.
- ✓ 16 Hours Instructor-Led Lab
- ✓ Full Python Code Repository Access
- ✓ Capstone Review
- ✓ Certificate of Completion
- ✓ HRD Corp Claimable
- ✓ Priority Mentoring Placement
- ✓ 16 Hours Instructor-Led Lab
- ✓ Full Python Code Repository Access
- ✓ Capstone Review
- ✓ Direct 1-on-1 Architecture Q&A
- ✓ HRD Corp Claimable
- ✓ Designed for Engineering Squads
- ✓ 16 Hours Instructor-Led Lab
- ✓ Corporate Consolidated Invoicing
- ✓ Post-Cohort Team Roadmap Review
- ✓ HRD Corp Claimable
Next batch · date to be announced.
Instructor-led sessions are intentionally kept small for hands-on mentoring.
Company paying? Ask for the course outline and HRD Corp quotation on WhatsApp →
Technical & logistics details
You need basic programming experience. Python is the language we use, and Module 2 refreshes what you need: classes and objects, functions, dictionaries and lists, JSON and external libraries. You don't need to be an expert, but this is not a no-code course.
Yes. You'll work in Python and Jupyter Notebook inside VS Code, with Pydantic, SQLite, ChromaDB and LangGraph. You build and run real code on your own machine.
Business-automation courses focus on no-code connectors and productivity prompts. This course focuses on the engineering of Agentic AI: integrating LLMs into existing software, structured output, memory, RAG, state, tools and graph-based workflows.
Python, Visual Studio Code, Jupyter Notebook, LLM API integration, Pydantic, SQLite, ChromaDB and LangGraph.
You'll complete 13 progressive exercises: a structured project-status tool, a context-aware support assistant, a memory assistant, an HR policy RAG assistant, a multi-turn shopping assistant and a project assistant with LLM-based tool selection. They lead into a capstone Agentic AI application that combines LLM, context, memory, RAG, state, tools and a LangGraph workflow.
Yes. The training is fully HRD Corp claimable under the SBL-Khas scheme. Our admissions coordinators will provide all required course outlines, trainer profiles, and official quotation documents for the grant application.
The date is being locked in and will be announced soon. Join the priority list and you'll be the first to know, along with early-bird seat access.
Ready to build your first real agent?
Be first in line when the next batch date drops. Join the priority list today.