Production ML and capstone
Take this if you have models that work on your machine and nothing anyone else can use.
What you’ll be able to do
You can serve a model behind an API, monitor it for drift and degradation, and take a project of your own from a scoped plan to a shipped, documented, presented system.
What it assumes
At least one substantial model you have built yourself, plus Python, Git and basic command-line comfort.
- Price
- ₹1,499
- Length
- 6 weeks
- Sessions
- 6
- Effort
- ~8 h/week · 48 h total
- For
- College · Working professional
Adds it to My courses on your dashboard so you can find it again. Payment is not open yet — nothing is charged.
What’s taught, in order
- 01
MLOps foundations
Reproducibility, experiment tracking, data and model versioning. The unglamorous strand that decides whether a result you got last month can be recovered at all.
- 02
Serving and deployment
Wrapping a model in an API, containerising it, and the latency and cost consequences of each choice. Ends with something another person can send a request to.
- 03
Monitoring and drift
What to measure once real inputs arrive, how distribution shift shows up before accuracy does, and the alert that tells you a model has quietly stopped working.
- 04
Capstone: scope and plan
Choosing a problem that is genuinely finishable, writing down what done means, and identifying the data before writing any code. Most failed capstones fail here.
- 05
Capstone: build
Two weeks of supervised building on your own system, end to end, with checkpoints that catch scope drift while there is still time to correct it.
- 06
Capstone: ship, document, present
Deploying it, writing documentation somebody else can follow, and presenting the work — including what did not work, which is the part that makes a portfolio credible.
What you build
Take one system of your own from scoped plan to deployed reality in six weeks: a written scope with an explicit definition of done, a working deployment somebody else can reach, monitoring that would tell you if it broke, documentation sufficient for a stranger to run it, and a presentation covering the design decisions and the things that failed.
Check it yourself against this
- A written scope, dated before the build, with an explicit definition of done.
- A deployment another person can reach and use without your help.
- Monitoring exists and you can show what it reports.
- Documentation is sufficient for a stranger to run the system from scratch.
- The presentation includes at least two things that did not work and what you did about them.
- You can state the running cost of your system and what would make it too expensive.
Then answer this
Comparing your scope document with what you actually shipped — where did they diverge, and what would you scope differently next time?
Two or three sentences, in your own words. If you cannot, the course is not finished — go back to the module it came from.
What this course does not do
This is the practices and the shipping discipline at the scale of one person's project, not enterprise platform engineering: Kubernetes at scale, multi-region infrastructure, on-call and compliance work are all beyond it. Your capstone will be a portfolio piece, not a product.
Where this leads
The Production and Capstone phase page sets out the six weeks and the capstone checkpoints in detail, which is worth reading before you choose a project, since scope is the thing that decides whether it finishes.
FAQ
- Can I bring a project I have already started?
- Yes, and it often works better. The scoping week will still make you write down what done means, which is usually the step a half-finished personal project skipped, and is why it is half-finished.
- Do I need to pay for hosting?
- No. Every deployment exercise has a free-tier path, and the cost module teaches you to compute what a paid deployment would cost rather than requiring you to incur it.
- Is six weeks enough for a capstone?
- It is if the scope is honest, which is exactly what the first capstone week exists to enforce. Most capstones that fail were three months of work attempted in six weeks, and the checkpoint structure is designed to catch that in week four rather than week six.
Other courses
This is a standalone course. It does not add weeks to the 78-week curriculum or change your roadmap. If you want the whole path instead, the full programme is ₹399/month · ₹2,000/year — see what it covers.