Most courses call something a capstone when it is really just a bigger assignment stuck at the end. IIT Kharagpur's Executive Post Graduate Certificate in Building AI Products, Systems & Services does not work that way. Here, the capstone is not a last-minute task added to the syllabus. It runs across the full six months, and it is the one place where everything you learned in every module gets put to the test on a real product you build yourself.
That is what a genuine real-world capstone is supposed to do; it pushes you to design a complete AI application from start to finish or build a go-to-market product proposal that could actually stand in front of real decision makers. This post looks at what makes this capstone feel real, how it is set up, and why having a live faculty defence at the end changes what you actually walk away with.
Turn your AI knowledge into practical experience with the Executive Post Graduate Certificate in Building AI Products, Systems & Services from IIT Kharagpur. Work on industry-relevant projects and gain the skills to build AI solutions that deliver real business impact.
What Makes a Capstone "Real-World" Instead of Theoretical
A theoretical capstone usually ends with a slide deck or a written recommendation that nobody actually has to act on. A real-world capstone ends with something that behaves like an actual product, with real constraints attached to it.
In this program, that difference shows up in three ways.
You pick a real problem instead of working off a case study someone else already packaged for you, so the ambiguity and the stakes are genuinely yours to deal with
You build a working prototype with live LLM integration instead of a mockup, so the system actually has to function, not just look like it does
You defend your decisions live, in front of people who can push back and question you, instead of quietly submitting a document that gets graded somewhere behind the scenes
How the Capstone Project Actually Works
The capstone is not a separate track running alongside the nine modules. It runs right through them, so whatever you learn in a module becomes a piece of the same product you have been building all along.
You start by picking a real problem from any domain you want, which means you are applying the opportunity discovery skills from Module 1 to your own idea, not someone else's case study
From there, you carry that same problem through design, prototyping, evaluation, and go-to-market, using each module's tools on it as you move along
By the final stretch, you are not stitching together a bunch of separate assignments, you are refining the one system you have been building since day one
The Six Deliverables You Build Along the Way
The capstone produces the same set of artifacts you have been building all along, except now all six connect to one product instead of sitting as six unrelated exercises.
An AI opportunity and feasibility brief that actually justifies why the problem is worth solving with AI
A full product specification with clearly defined AI behaviour
A working prototype with live LLM integration that genuinely runs
A RAG and agentic system architecture for any part of your product that needs retrieval or multi-step reasoning
An evaluation and red teaming framework that tests your own system for bias and where it might fail
A business case covering both pricing and compliance
Put together, these six pieces tell one complete product story, not a summary written after the fact.
The Live Faculty Defence: Where the Pressure Testing Happens
The defining moment of the capstone is the live defence in front of an IIT Kharagpur faculty panel. This is not a casual Q&A where you click through slides.
You are expected to justify every meaningful decision behind your product, from why you picked a certain architecture to how you priced the feature
The panel throws in a real-time complication, like sudden model drift, and expects you to respond right there on the spot
This matters because it mirrors what actually happens after a product launches, no AI product ships and then just behaves perfectly forever
Models drift, edge cases show up, and users find ways to break things nobody planned for
Being pushed to think through that live, instead of writing about it later in a calm retrospective, is what makes this feel closer to a real product incident than a classroom exercise
Example Capstone Directions Learners Can Choose
Since the capstone starts with a problem, you pick yourself, the directions people take end up looking very different depending on their industry and role.
Some build a customer support copilot that cuts down ticket resolution time by pulling answers from a company's own documentation
Others design an internal analytics assistant that lets non-technical teams ask questions about business data in plain language
Founders often use the capstone to prototype the exact AI feature they are planning to launch in their own product
Professionals in operations or finance tend to build fraud or anomaly detection systems tied to their own sector
The direction you pick matters less than the discipline you bring to it. Whatever problem you choose, you are still expected to justify it, build it, evaluate it, and defend it with the same level of rigour.
Why This Capstone Format Matters for Your Portfolio
A written proposal tells an employer you can think about AI products. A working prototype backed by a defended architecture, an evaluation framework, and a business case tells them you can actually deliver one.
That gap matters more now than it used to, most hiring managers have already seen plenty of AI themed slide decks
What they are actually looking for now is someone who can show, not just describe
Because the IIT Kharagpur capstone produces real artifacts tied to a real problem, you end up with something you can walk a hiring panel or a leadership team through directly
You can use the same reasoning you already defended in front of the IIT Kharagpur faculty panel
Conclusion
The capstone in IIT Kharagpur's Building AI Products, Systems & Services program is meant to feel uncomfortable, the same way a real product launch feels uncomfortable. You pick the problem, you build the system, and you defend it live with no script to fall back on. That is what turns six months of modules into one solid proof point, not that you understand AI products, but that you can actually build one.