Certification programme
The newest thing we teach, and the one nothing else in Mumbai covers properly. Building AI systems that plan, call tools and finish multi-step work on their own.
What you will be able to do
Full syllabus
8 modules · 40 topics · 159 concepts
Definitions
When agents are the wrong answer
LLM fundamentals for builders
Structured output
The development setup
You can tell whether a problem needs an agent or just a prompt.
Prompt structure
Techniques
Reliability
Versioning and testing
Cost and latency
You can write and regression-test a prompt like a function.
Embeddings
Chunking
Vector stores
Retrieval quality
Memory
You can build a retrieval system and measure whether it actually retrieves.
Function calling
Designing tools
Connecting real systems
Failure handling
Permissions
You can give a model tools and constrain what it is able to do with them.
Single-agent loops
Graph-based orchestration
Multi-agent systems
Human in the loop
Cost and depth control
You can choose and implement an architecture appropriate to the task.
Why evaluation is hard
Building a test set
Evaluation methods
Metrics
Observability
You can build an evaluation suite that catches regressions before users do.
Prompt injection
Data exfiltration
Tool abuse
Output safety
Operational safety
You can identify and mitigate the ways an agent can be made to misbehave.
Serving
Cost management
Reliability
Deployment
Product judgement
You can deploy an agent with proper controls, monitoring and cost management.
5 builds you keep
Multi-step search and synthesis where every claim in the output must trace to a retrieved source. Graded on whether unsupported statements appear, not on how good the prose is.
Reads tickets, proposes an action, and waits for a human before anything irreversible. Demonstrates interruption, resumption and the audit log that makes it acceptable to deploy.
You are given a working agent and asked to make it leak data or take an action it should refuse, using indirect prompt injection through documents it retrieves. Then you defend it and re-test.
Build the test set, the assertions and the scoring, then use it to catch a real regression when a prompt or model version changes. The unglamorous piece that decides whether the rest is real.
A working agent behind an API with per-user spend caps, tracing, an evaluation suite running in CI, and a documented list of what it is not permitted to do.
Hands on with
Before you enrol
Comfortable Python and prior experience calling an LLM API. This course does not teach programming or how a transformer works. Our AI course is the right prerequisite if you need that background.
Partly, and the first module says so directly. Plenty of things labelled agents are prompt chains with a marketing budget. The genuinely useful cases are narrow and tool-heavy, and the difference is exactly what this course is about.
Primarily LangGraph, because explicit graphs are more debuggable than free-form loops, and we build the core loop without a framework first so you know what the framework is doing. Frameworks in this space change fast; the patterns do not.
They are manageable, and cost control is taught from the first module rather than as an afterthought: budget caps, caching and routing simple sub-tasks to smaller models. We show you how to keep a project inside a small monthly spend.
Titles vary a lot right now: AI engineer, applied AI developer, and increasingly ordinary backend roles where an AI feature is part of the work. The evaluation and security modules are what distinguish candidates, because most applicants have only built a demo.
Where this leads
One skill on a CV is a hobby. These are the courses students take alongside this one, in the order that builds an actual role.
Run real campaigns with real budgets, with AI doing the production work. SEO, paid ads, content and analytics, measured properly so you can prove what worked.
Defensive security aimed where the attacks now are: cloud infrastructure and AI systems. Everything is practised in an isolated lab we provide.
Deploying is the easy part. This is the course about what happens afterwards: Kubernetes in depth, GitOps, observability you can debug with, and the reliability practices that decide whether a system stays up at three in the morning.