Programs/AI Engineering

Generative AI & Agentic AI

From intelligent answers to controlled, useful actions.

Build, evaluate and deploy LLM-powered applications and agents. Guided labs and capstone work are included in the course hours.

108

Learning hours

Connected modules

Cloud capstone

Book a Free Consultation See the Curriculum

Your build path

01
Understand
Models + prompts
02
Ground
Documents + retrieval
03
Act
Tools + approval
04
Deploy
Evaluate + operate

The Opportunity

What will you learn to build?

Move from isolated prompts to applications that retrieve evidence, call tools and run within clear boundaries.

Generative AI

Creates answers, summaries, code or structured content from instructions and context.

Example: Summarize a support document.

Agentic AI

Uses a model to choose steps and call tools toward a goal.

Example: Retrieve a policy, inspect a ticket and propose a next action.

Why learn both?

Turn language-model capabilities into useful software with APIs, state and reliable error handling.
Ground answers in your own documents and evaluate whether the evidence supports them.
Design tool-using workflows with permissions, stop conditions and human approval.

Is this programme for you?

Designed for learners with Python and introductory machine learning.

Suitable for developers and data professionals building AI application skills.

No prior agent-building experience is assumed. This course develops practical application engineering; it does not promise mastery of every model or platform.

How It Builds

5 learning phases, one connected project

Each phase adds a layer to the same application — from model calls to a deployed, evaluated agent.

Phase 01 · 30 hours

Understand models. Design clear instructions.

Start with how language models work, then learn to select and direct them for a specific task.

Module 01 · 8 hours

Introduction to Machine Learning

Understand the fundamentals of Machine Learning, how machines learn from data, and the different types of ML models. Explore real-world applications, basic workflows, and how ML is shaping modern AI and technology.

Practice: Introduction to Machine Learning.

Module 02 · 8 hours

Generative AI & LLM foundations

Tokens, embeddings, attention, Transformers, pretraining, alignment, context windows, inference and model limitations.

Practice: Explain why an answer can be fluent but incorrect.

Module 03 · 6 hours

Model selection & access

Base, instruction, reasoning and multimodal models; embeddings and rerankers; APIs versus local inference; cost and latency.

Practice: Compare model choices against a task and budget.

Module 04 · 8 hours

Prompting & structured outputs

Zero-shot and few-shot prompts, message roles, context construction, task decomposition, JSON schemas, validation and prompt tests.

Practice: Create and validate a structured support-ticket response.

Milestone: A repeatable model call, a tested prompt and a validated output contract.

Phase 02 · 24 hours

Ground models. Engineer the application.

Make source content searchable and connect the model to an application workflow.

Module 05 · 14 hours

Embeddings, vector databases & RAG

Ingestion, chunking, vector search, metadata filters, hybrid retrieval, reranking, citations, retrieval quality and answer evaluation.

Practice: Build document question-answering with source references.

Module 06 · 10 hours

LLM application engineering

SDK integration, LangChain, LangGraph, stateful workflows, persistence, streaming, async execution, retries, timeouts and caching.

Practice: Expose the workflow with predictable error handling.

The RAG pattern

Prepare
Parse and chunk
Retrieve
Select evidence
Generate
Answer with sources

RAG supplies relevant evidence to the model. Keep citations, test retrieval and return a clear no-answer response when the source material is insufficient.

Phase 03 · 22 hours

Give agents tools. Keep actions controlled.

An agent needs more than a prompt: define its tools, state, permissions and stopping rules.

Module 07 · 12 hours

Tool use, planning & memory

Tool calling, agent loops, planning, task decomposition, memory, stop conditions, error recovery and human in the loop.

Practice: Build a bounded tool-using support workflow.

Module 08 · 10 hours

MCP, integrations & multi-agent systems

MCP host/client/server, tools, resources and prompts; API and database integration; permissions, handoffs and supervisor patterns.

Practice: Connect an external capability and document its permissions.

A controlled agent loop

Decide
Choose the next step
Validate
Check permission
Execute
Call the approved tool

Observe the result and continue only within the step/time budget. Require human approval for sensitive actions; stop or escalate when the task cannot be completed safely.

Phase 04 · 18 hours

Improve quality. Measure what changed.

Use a test set to compare decisions rather than relying on a few impressive demonstrations.

Module 09 · 10 hours

Fine-tuning & model optimization

Dataset preparation, supervised fine-tuning, PEFT, LoRA, QLoRA, quantization, validation and quality-memory-latency trade-offs.

Practice: Compare an adapted or optimized model with a baseline.

Module 10 · 8 hours

Evaluation, safety & observability

Golden datasets, groundedness, LLM-as-judge, tool correctness, regression tests, prompt injection, PII, access controls and tracing.

Practice: Track failures and compare a new version with the baseline.

Evaluate four dimensions

Answer quality
Evidence + relevance
Action quality
Correct tool + arguments
Reliability
Latency + error handling
Operating cost
Usage + resource spend

Phase 05 · 14 hours

Integrate the system. Deploy on one cloud.

Choose Azure or AWS for the capstone deployment, according to the batch delivery plan.

Module 11 · 6 hours

AI system design & capstone integration

Architecture decisions, component integration, end-to-end testing, documentation and preparation for cloud deployment.

Practice: Connect retrieval, tools, state, approval and evaluation.

Module 12 · 8 hours

Cloud deployment of AI agents

Cloud model access, identity and secrets, FastAPI, Docker, retrieval and state integration, logging, cost controls and scaling.

Practice: Deploy and operate the enterprise support agent.

Your application stack

Build & orchestrate
Python • SDKs • LangChain • LangGraph • FastAPI
Deploy & operate
Docker • Azure or AWS • identity • secrets • logs • state

Vector store and model choices depend on the lab environment. Cloud accounts, model access and usage charges should be confirmed before enrollment.

Curriculum · 108 Hours

12 modules. One continuous build.

All guided labs and capstone work fit within these hours.

Phase 01 · Understand models. Design clear instructions.

Practice: Introduction to Machine Learning.

Phase 01 · Understand models. Design clear instructions.

Tokens, embeddings, attention, Transformers, pretraining, alignment, context windows, inference and model limitations.

Practice: Explain why an answer can be fluent but incorrect.

Phase 01 · Understand models. Design clear instructions.

Base, instruction, reasoning and multimodal models; embeddings and rerankers; APIs versus local inference; cost and latency.

Practice: Compare model choices against a task and budget.

Phase 01 · Understand models. Design clear instructions.

Zero-shot and few-shot prompts, message roles, context construction, task decomposition, JSON schemas, validation and prompt tests.

Practice: Create and validate a structured support-ticket response.

Phase 02 · Ground models. Engineer the application.

Ingestion, chunking, vector search, metadata filters, hybrid retrieval, reranking, citations, retrieval quality and answer evaluation.

Practice: Build document question-answering with source references.

Phase 02 · Ground models. Engineer the application.

SDK integration, LangChain, LangGraph, stateful workflows, persistence, streaming, async execution, retries, timeouts and caching.

Practice: Expose the workflow with predictable error handling.

Phase 03 · Give agents tools. Keep actions controlled.

Tool calling, agent loops, planning, task decomposition, memory, stop conditions, error recovery and human in the loop.

Practice: Build a bounded tool-using support workflow.

Phase 03 · Give agents tools. Keep actions controlled.

MCP host/client/server, tools, resources and prompts; API and database integration; permissions, handoffs and supervisor patterns.

Practice: Connect an external capability and document its permissions.

Phase 04 · Improve quality. Measure what changed.

Dataset preparation, supervised fine-tuning, PEFT, LoRA, QLoRA, quantization, validation and quality-memory-latency trade-offs.

Practice: Compare an adapted or optimized model with a baseline.

Phase 04 · Improve quality. Measure what changed.

Golden datasets, groundedness, LLM-as-judge, tool correctness, regression tests, prompt injection, PII, access controls and tracing.

Practice: Track failures and compare a new version with the baseline.

Phase 05 · Integrate the system. Deploy on one cloud.

Architecture decisions, component integration, end-to-end testing, documentation and preparation for cloud deployment.

Practice: Connect retrieval, tools, state, approval and evaluation.

Phase 05 · Integrate the system. Deploy on one cloud.

Cloud model access, identity and secrets, FastAPI, Docker, retrieval and state integration, logging, cost controls and scaling.

Practice: Deploy and operate the enterprise support agent.

Total learning hours108h

Build the enterprise support agent progressively: AI system design brings it together, and the final module deploys and operates it on one platform — Azure or AWS.

Capstone · Built Throughout the Course

Enterprise support agent

One complete project connects all 12 modules. Guided implementation and deployment are included in the course hours.

01
Knowledge
Documents and policies
02
Retrieval
Relevant source evidence
03
Model
Grounded response
04
Agent workflow
Tools, state and approval
05
Cloud application
Authenticated API + logs

Example task

Find a policy, inspect a ticket and propose the next action for approval.

What you will present

Hosted application and authenticated API.
Document answers with citations, tool access and conversation state.
Human-approval flow, architecture diagram and deployment guide.
Quality, latency and cost report with test results.

A portfolio project should show both successful scenarios and known limitations. Use permitted or synthetic data for demonstrations.

Career Path

Show your skills. Explain your decisions.

Roles to explore

AI Application DeveloperGenAI EngineerAgent Developer

Fit depends on prior experience, coding ability and employer requirements.

Interview evidence

Explain model choice, chunking, retrieval, tool permissions, failure recovery and evaluation. Demonstrate your deployed capstone.

Career guidance: discuss resume review, mock interviews and opportunity guidance with J2D. Confirm scope and eligibility; employment is not guaranteed.

Start your AI engineering journey.

Ask about prerequisites, batch schedule, delivery format, fees and cloud lab access.

Book a Free Consultation Call +91 99646 13666

Also see: Data + GenAI Engineer