Included with a plan

Direct the analyst. Get graded on the instruction.

13 sessions against real data. You do not write code. You write the instruction, an AI analyst runs it in your own sandbox, and each step is graded on whether the instruction you wrote was actually complete.

It is the closest thing on the site to the job itself, which is why it is also the fastest way to find out which part of the work you cannot yet describe precisely.

13 sessions130 graded steps11 hours of practice0 code writtenself-paced

The instruction you wrote is the thing being marked, not the answer the model returned.

How a session runs

Four steps, repeated until the instruction holds.

Read the scenario

A real business situation with real data behind it. No toy dataset, and no single obvious answer.

Write the instruction

You brief the analyst the way you would brief a person: what to do, on what, and what a good answer looks like.

Watch it run

The analyst executes in your own sandbox. Where your brief was vague, the output shows it immediately.

Get graded, then tighten it

Each step is marked on whether the instruction was complete, so the feedback is about your thinking, not the model's mood.

Every session, in the open

All 13, with the scenario each one runs.

Nothing is hidden behind the plan except the sandbox itself. Open any session below to see what it puts in front of you.

S1AI Product Teardown10 steps, 50 min

The Chaayos Recce

You finish with: By the end you will have shown, with real numbers and a rupee figure, that paying 99% of invoices is not the same as paying them correctly, and you will have a short teardown note saved in your workspace.

S2Discovery at 10x Speed10 steps, 50 min

The Complaint Goldmine

You finish with: By the end you will have found the real, best-evidenced problem to solve, caught a made-up theme that has no evidence behind it, and saved a discovery method you can reuse on any future problem.

S3Model and Data Feasibility10 steps, 50 min

The Three Apprentices

You finish with: By the end you will have a clear recommendation on how to build it (prompt, retrieval or fine-tuning), a data readiness check that names the gaps no model can fix, and a short feasibility note saved in your workspace.

S4The AI Opportunity Brief10 steps, 50 min

Uncle Mahesh's Two Pages

You finish with: By the end you will have a rupee figure for the prize with every assumption named, a risk register with owners, a read on where the product sits under the rules, and a two-page brief a sceptical exec could sign.

S5Context Is the Spec10 steps, 50 min

The Card on the Wall

You finish with: By the end you will have a versioned context spec with clear rules and refusals, a retrieval design that cites its sources and handles stale documents, and a break log of the failures you found first.

S6Agents, Tools and Orchestration10 steps, 50 min

Chintu Gets a Phone

You finish with: By the end you will have a clear agent plan, two or more tool definitions grounded in the real data, a working approval gate that catches a high-value trap, a memory rule, and a written handoff.

S7Vibe-Code the Prototype10 steps, 50 min

The WhatsApp Weekend

You finish with: By the end you will have a tightly scoped prototype whose three requests are proven on real data, a clear line on what you refused to build, a stranger test, and a short reproducibility note.

S8Design for Trust and Failure10 steps, 50 min

The Prankster Invitation

You finish with: By the end you will have three designed sad paths, a set of guardrail patterns, a red-team log of what you tried and what you fixed, and a bias spot-check on a matched pair.

S9Evals, Prove It Works10 steps, 50 min

The Blind Taste Test

You finish with: By the end you will have a named failure taxonomy, an honest read of the eval set, an understanding of the AI-judge blind spot and human agreement, and a before-and-after plan that reports a real delta.

S10Evals in Production10 steps, 50 min

Winter Comes to the Tapri

You finish with: By the end you will have caught the drift day from the data with evidence, a clear tracing plan, a regression-gate rule, and an incident playbook with detect, act, and fallback.

S11Metrics, Cost and Unit Economics10 steps, 50 min

The Price of One Cup

You finish with: By the end you will know the cost per task and how it compares to a human, the saving from one cost lever, a fuller total cost, and a valid experiment plan with a guardrail.

S12The AI PRD and Business Case10 steps, 50 min

The Family Panchayat

You finish with: By the end you will have an AI PRD with the four extra sections, measurable kill criteria, a staged rollout, a five-slide case, and honest answers or written-down open risks for all four objections.

S13Ship the Portfolio and Demo Day10 steps, 50 min

The Tech Park Audition

You finish with: By the end you will have your artefact chain assembled, a four-beat interview narrative, the five interview questions pre-answered, a three-minute demo script, and an honest capstone self-score.

Before you buy a plan for this

Who it works for, and who it does not.

Worth your time if

this is the gap you actually have

You can talk about AI but have never directed it against real data
Your interview answers go vague the moment someone asks what you did
You want reps, on your own schedule, with honest marking

Thirteen sessions is a few evenings, not a term.

Not the right thing if

something else on the site fits better

You want live coaching and a cohort: take the boot camp instead
You have an interview this week: run 1 Day Prep in the toolkit
You only want to read questions: the bank is free and needs no plan

All three are linked below.