Hard prerequisites
Python proficiency — functions, classes, HTTP requests, working with JSON. Familiarity with REST APIs — you should know what a POST request is and have made at least one API call in code.
// FLAGSHIP PROGRAM
Generative AI and Agentic Systems
Go beyond using AI tools — learn to build the systems that power them.
6 weeks · Intensive, project-driven
In-Person
Online
Hybrid
For builders, not just users
There is a difference between a person who uses AI tools and a person who builds AI systems. The first uses a chatbot. The second builds one — and knows why it behaves the way it does, how to make it better, and how to stop it from going wrong.
This program is for the second person.
Over six intensive weeks you move through the full stack of modern generative AI — how large language models work, how to engineer prompts that produce consistent and reliable output, how to build multi-step LLM workflows, how to design agentic systems where AI takes sequences of actions autonomously, and how to ground AI output in real data using retrieval-augmented generation. Every concept is built and tested with working code. You leave with systems you built yourself — deployed, documented, and defensible.
This is not a course about using ChatGPT more effectively. It is a course about building the infrastructure that makes AI useful in production.
For technical learners ready to build production AI features
This program is designed for:
- Developers who have completed the AI-Native Full Stack program (or equivalent Python and API experience) and want to specialise in AI systems
- ML practitioners who understand model training but want to go deeper into LLM application development
- Working professionals in technical roles — developers, analysts, data engineers — who need to build AI-powered features and workflows
- Ambitious students from any technical background who want to position themselves at the frontier of what employers are hiring for
Recommended background: you should be comfortable with Python, REST APIs, and have a working understanding of how to call an external API. You do not need ML or deep learning experience — this program works with LLMs as capabilities, not as models you train.
If you have completed the AI-Native Full Stack program, you are well-prepared. If you have not, speak to us and we will assess your readiness honestly.
Ready for an intensive six weeks
Soft prerequisites
Some exposure to AI tools as a user — you should have used ChatGPT, Claude, or similar tools and formed opinions about what they do and do not do well. Curiosity about failure modes is an advantage.
Setup before day one
Python 3.10+, VS Code, Git, a GitHub account, and API accounts for OpenAI and Anthropic (both have free tiers). A full setup checklist is sent after enrollment. Exercises are designed to stay within free-tier limits wherever possible, and we will tell you before anything incurs cost.
Hardware
Any modern laptop with stable internet is sufficient for the course. Most work is API-based, not GPU training.
Module by module
Seven modules, designed to take you from foundations to production deployment. Every module is implemented with working code and tested for real failure modes.
Understand what is happening inside LLMs so you can predict failure modes and build systems that behave reliably.
Topics
- Transformers, tokens, context windows, pre-training vs instruction tuning, RLHF
- Temperature, top-p, sampling, and why hallucination is structural
- Model landscape: GPT-4o, Claude, Gemini, open-source options and trade-offs
- Capability vs reliability gap, and selecting the right model for a task
Hands-on
- Structured capability testing across multiple models, with documented failure analysis
Prompting as an engineering discipline: systematic, testable, and versioned like code.
Topics
- Zero-shot, few-shot, chain-of-thought, system prompts, structured output and validation
- ReAct prompting, prompt chaining, evaluation, version control for prompts
- Prompt injection and defense patterns
Hands-on
- Build a prompt test suite for a real business task, then break it with injection attempts and harden it
Build multi-step LLM workflows with memory, document handling, and structured outputs — and learn when not to use a framework.
Topics
- Chains, prompt templates, output parsers, sequential pipelines
- Memory patterns: buffer, summary, vector, entity
- Document loaders and chunking strategy, plus “when not to use LangChain”
Hands-on
- Document Q&A system with conversational memory; debug and fix a failure caused by poor chunking
Build a production-grade RAG pipeline from embeddings to evaluation, including real-world failure modes tutorials skip.
Topics
- Embeddings, similarity metrics, vector stores (FAISS, Chroma), metadata filtering
- Hybrid retrieval, reranking, query transformation, context stuffing issues
- Evaluation, including RAGAS framework introduction
Hands-on
- Embed a real document set, implement retrieval, then deliberately break it and rebuild with better chunking and measured improvement
Build agents with explicit state and control flow: tool use, loops, human-in-the-loop checkpoints, and multi-agent orchestration.
Topics
- Agent loop, tool schemas, ReAct, failure modes, agent vs chain decisions
- LangGraph: nodes, edges, state schema, conditional routing, cycles, persistence
- Tool design and security, orchestrator-worker patterns, failure handling
Hands-on
- Research agent with citations and confidence, code review agent, and human-in-the-loop action checkpoints
Ship AI systems safely: prompt injection defense, output validation, DPDP awareness, cost controls, monitoring, and versioning.
Topics
- Prompt injection, jailbreaking, data exfiltration, output sanitisation, rate limiting, secret management
- Responsible AI design, privacy and PII handling, India DPDP Act basics, transparency and refusal patterns
- Deployment: FastAPI wrappers, streaming and caching, monitoring, prompt and model versioning
Hands-on
- Security audit of previous systems, implement fixes, deploy a complete AI application with monitoring and cost alerts
Build a complete deployed agentic or RAG-powered system, scoped with your trainer, and defend it in a viva like a real AI engineer interview.
Example systems
- Legal: contract intelligence agent with multi-step extraction, RAG over clause library, validation and disclaimers.
- Enterprise knowledge: private knowledge assistant with citations, confidence thresholds, and human escalation.
- Software: autonomous code assistant with sandbox execution and human-in-the-loop at PR generation.
- Customer experience: tier-1 support agent with tools, escalation summaries, and session logging.
- Research: multi-source research agent with contradiction checks, citations, and confidence assessment.
Assessment
25-minute viva — demo the deployed system, walk through architecture, explain design decisions, describe failure modes and mitigations, and answer questions on scale.
A deployed system you can defend
The course ends with a portfolio-grade build: deployed, documented, secure, and interview-ready.
Your project is jointly scoped with your trainer in a structured conversation. You choose a domain. Together you design a system that is achievable in six weeks, technically demonstrates the core skills of the course, and is impressive without being impossible.
What you leave with
- One deployed agentic or RAG-powered application with a public URL
- GitHub repository with prompts, evaluation notes, and documentation
- Security hardening: injection defense, output validation, and secret management
- Viva experience defending architecture, failure modes, and trade-offs
Stack you will use in every module
LLM APIs
OpenAI API (GPT-4o, embeddings, function calling) · Anthropic Claude API · Google Gemini API (awareness)
Frameworks
LangChain · LangGraph · LangSmith (observability)
Vector stores
FAISS · Chroma · Pinecone (awareness)
Deployment
FastAPI · Streamlit · Render · Railway · Hugging Face Spaces
Development environment
VS Code · Jupyter Notebook · Git · GitHub · Postman
Security and observability
python-dotenv · LangSmith tracing · Weights and Biases (awareness)
Roles this program prepares you for
AI Application Developer
LLM Engineer
Generative AI Engineer
AI Solutions Engineer
AI Product Developer
ML Engineer (NLP / LLM systems)
Prompt Engineer (senior level)
Why this program is commercially distinctive: the ability to build agentic systems — AI that takes sequences of actions, uses tools, and operates with partial autonomy — is one of the most sought-after and least commonly available skills in the Indian market in 2026.
There are many people who can use ChatGPT. There are very few who can build a LangGraph agent with RAG, tool use, human-in-the-loop, and production deployment. This program produces the second profile.
Companies hiring these profiles from Tamil Nadu and beyond: Freshworks, Zoho, and Chargebee have active AI engineering teams in Chennai. Wipro, TCS, and Infosys have AI labs hiring for LLM application work. GCCs in Chennai — Standard Chartered, HSBC Technology, DBS, Maersk — are building internal AI systems. Remote-first AI product companies globally hire this profile from India actively.
What you walk in with vs. what you walk out with
| Walk in with | Walk out with |
|---|---|
| Python and API experience | Full LLM application development capability |
| Basic AI tool use | Prompt engineering as a systematic, testable discipline |
| No LangChain knowledge | LangChain and LangGraph agentic system architecture |
| No RAG experience | Production-grade RAG pipeline with evaluation methodology |
| No agent design experience | Multi-step LangGraph agents with tool use and human-in-the-loop |
| No AI security knowledge | Prompt injection defense, output sanitisation, DPDP Act awareness |
| No deployed AI systems | Complete deployed AI application — live URL, monitoring, documented |
Three modes, one standard
In-Person
In-Person — Tiruvallur campus: daily sessions, live coding alongside trainer, immediate help on API integration issues and debugging. The agentic modules particularly benefit from in-person collaboration — debugging agent loops is easier with a second pair of eyes in the room.
Online
Online: live instructor-led sessions via Zoom or Google Meet. Same curriculum, same projects, same assessment. Screen sharing and collaborative debugging work well for this program online.
Hybrid
Hybrid: attend in-person for the denser modules (RAG and agents especially), online for the others. Recordings available for Online and Hybrid students for review — not as a replacement for live attendance.
All three modes deliver the same curriculum and the same assessment. No mode is a reduced version.
// Enquire
Ready to find out when the next batch starts?
Call us: 63851-58458 | 98409-41910 — or use the button below and we will call you within 24 hours.
