IIT Roorkee Advanced Certificate Programme in Agentic AI and Generative Systems Engineering

• Introduction to ML (High Level), AI vs ML vs DL, bias-variance tradeoff

• Feed-forward Neural Network, Backpropagation and Gradient Descent, variants of gradient descent

• CNN, image classification

• Basic Text Processing, TF-IDF, Sequence Labelling Task (POS Tagging, HMM), syntactic parsing, dependency parsing

• RNN, LSTM, attention mechanism, sequence-to-sequence learning

• Sentiment analysis, Machine Translation, Information extraction and question answering

• Transformer architecture, self, multi-head and cross attention

• Introduction to language models, large language models, encoder-only, decoder-only, encoder-decoder

• Pretraining and fine-tuning of LLMs: hands-on

- Generative AI Basics: what is Generative AI; applications (chatbots, image generation)

• AE, VAE, GANs, Diffusion Models

• GenAI Tools + Mini Projects: ChatGPT, DALL·E, Azure AI Studio; build a text generator; build an image-based app

- Prompt Engineering: prompt design, few-shot prompting, chain-of-thought reasoning; practice on real tasks

• Multimodal foundation models (GPT-4o, Gemini, Claude, Llama Vision)

• Vision APIs, OCR and document understanding

• Multimodal embeddings and retrieval concepts

• Multimodal prompting and function calling

• Building multimodal AI applications (image/document assistant)

- Agentic AI Fundamentals: what is Agentic AI; agent lifecycle; decision loops

- Agent Architecture: perception → reasoning → action; tools & memory; planning

- LangChain: chains, agents, tools & memory; build a simple agent

- LangGraph: nodes & routing; workflow orchestration; build multi-step pipelines

- Multi-Agent Systems: single vs multi-agent; collaboration patterns; task decomposition

- AutoGen & CrewAI: multi-agent frameworks; agent communication; build agent teams

- MCP (Model Context Protocol): basics; messaging & tool integration; build MCP-based workflow

- RAG Basics: Retrieval-Augmented Generation; embeddings; vector DB concepts

- Agentic RAG: combine agents + RAG; external APIs; build domain-specific assistant

- AI Trust & Responsible AI: explainability and transparency concepts; bias, fairness and responsible AI principles; human-in-the-loop validation; AI governance fundamentals

- LLM & Agent Security: prompt injection attacks; data leakage protection; secure tool usage and agent guardrails; risks in RAG and multi-agent systems

- AI Risk Management & Monitoring: model evaluation and red teaming basics; hallucination detection and mitigation; AI monitoring, observability and audit logs; enterprise AI TRiSM best practices

- Capstone Project: design a multi-agent system; design MCP + LangChain / LangGraph; start implementation

- Capstone Completion: finalize project; testing & evaluation; documentation (README, architecture); demo-ready system

• Python and AI-assisted coding assignments; prompt engineering exercises.

• Hands-on projects using LLMs and foundation models; GenAI mini-projects (text and image generation).

- Multimodal Gen AI projects: image understanding, OCR, document intelligence, speech.

• Agentic AI assignments on planning, reasoning, memory and tool integration.

• LangChain and LangGraph workflow implementation; Multi-Agent projects using CrewAI and AutoGen.

• MCP-based agent orchestration; Agentic RAG projects using embeddings and vector databases.

• AI security and governance case studies (prompt injection, hallucination mitigation, risk management).

- Capstone: design and build a production-ready Agentic AI / Multimodal / Agentic RAG solution with AI TRiSM considerations.