Skip to content
NAADI AI
How It's Built

Every Decision Here Has A Reason.

This page is the reasons. The mechanics, the numbers, and the four features we built and then deleted — because a product you can check is worth more than one you have to trust.

No predicted ranks. No guaranteed scores. Nothing on this page that we cannot show you inside the product.

What we build on

Built around how memory actually behaves.

Four findings decide almost everything else on this page. None of them are ours — they are the parts of the learning research that survived replication.

  • Retrieve, then revise. Pulling an answer out of your head builds memory. Reading the same page again mostly builds the feeling of knowing it.
  • Correct the concept, not the chapter. "Revise Thermodynamics" is not a plan. The gap is one idea inside it, and that is the unit we track.
  • A path that adapts. What comes next depends on what just happened. A fixed syllabus cannot do this, which is why every student here gets a different sequence.
  • Progress that never falls. Effort already spent is not taken back by a bad day. Mastery per concept can drop; the record of work done cannot.

The limit: research tells you the shape of a good system, not the settings. Every number on the rest of this page had to be found by testing, not by reading.

Where the questions come from

A chapter on one side. Every past paper on the other.

Questions are what is left when the two are made to agree. Here is one real block of NCERT, walked all the way through.

Hydrocarbons · NCERT Class 11 · p. 3

"Alkanes are saturated open chain hydrocarbons containing carbon-carbon single bonds."

A block is two or three lines carrying exactly one testable thing — a definition, a formula, an exception, a comparison, an ordered process, a named value. Not a paragraph. Not a topic. Every block is then scored for exam weight and matched against how past papers have actually treated it.

What we don't publish: the weighting, the trap categories, and the rules that turn one question into its variants. We publish what they produce, and you can check that against a real paper — which is what section nine does.

Concept Studio

One concept. Five ways in.

The same idea, written five times for five moments: the first read, the practice, the exam hall, the revision, and the proof.

01
Know it exactly

The definition in the wording NEET quotes, the one-line version to memorise, and the NCERT page, tagged.

02
Apply it

A worked scenario with real numbers, the cue that this concept is the one being tested, and what most often goes wrong.

03
Anticipate the examiner

How the question will be phrased, the neighbouring concept it gets confused with, and an assertion–reason drill on the exact pair.

04
Keep it

Say it back without looking, then flashcards and a concept map showing what this idea connects to on either side.

05
Check it against a real paper

This concept was asked in NEET 2025 — the real question, plus five practice variants built from it. Not a claim that it matters. The paper it appeared in.

Adaptive Practice · OPD™

Getting it right is the start. Keeping it is the point.

So a concept you have just fixed comes back — once soon, and once after it has had time to fade. The second visit is the one that tells you it stayed.

First ask. You get it wrong. You read exactly why — including why the option you chose was the tempting one. v1
Next test. Same concept, different wording — same answer. One test later, not immediately. Immediately is a memory test. A week later is a new lesson. v2 · T+1
Three tests later. Different wording — and a different answer. Trapped differently, with a different key, so getting it right can only mean the idea itself stayed. v3 · T+3
Wrong twice — the test stops.

No third strike in silence. The tutor names the misconception, walks every option and why yours was tempting, then gives a differently-trapped version before the test continues.

Right — and we come back to it.

Answering correctly just after reading the explanation can mean the rule landed, or that the answer was simply still fresh. Today those look the same. So we ask again once it has had time to fade.

If the later check doesn't hold, the concept simply reopens. Nothing is deducted and nothing is held against you — it comes back around, because that is what it was there to find out. It is also the number the teacher and parent portals report as retention.

One record, per concept

Two students, one chapter, two different tests.

Every question carries the concept it tests, so every answer writes to that concept — not to the chapter. Once that record exists, no two journeys can stay the same.

What gets written

Alkane nomenclature mastered78%
Electrophilic substitution learning66%
Alkyl groups struggling29%

Three concepts inside one chapter, three states, three different next steps. Averaged into a single chapter score they read 58% — a number that tells you nothing about what to open tonight.

What it changes

Student A— two concepts owed, one check due6 new · 3 back · 1 check
Student B— clean run so far9 new · 1 back · 0 checks

Same chapter, same evening, same twenty minutes. Neither test exists until it is asked for: the questions owed, the concepts due a check and the difficulty mix are read off that student's own record at the moment they press start. A struggling concept comes back sooner and easier; a mastered one steps up or steps aside.

The honest limit: this needs evidence to work. On the first test the record is empty, so everyone starts in much the same place. The paths separate as the record fills — which is why the early tests are short and frequent rather than long and rare.

Where every path is heading

The mix moves toward the real paper.

However your route through a chapter is built, it climbs the same six phases — each one longer and harder, until the test in front of you is shaped like the exam.

Foundationall easy10 Q · 8 min
Skill Building50 easy · 50 medium12 Q · 10 min
Mastery30 · 50 · 20 hard15 Q · 12 min
NEET Simulation60 medium · 40 hard, no easy20 Q · 15 min
Grand Mockcapstone, clean25 Q · 20 min
Endurancewhat the chapter has left20 Q · 15 min
By the last full test before the capstone there are no easy questions left in the mix at all — 60% medium, 40% hard, which is roughly what the paper itself does.

The Grand Mock then carries no remediation content whatsoever: no questions you got wrong, no checks, nothing owed. A capstone with your own corrections folded back into it is not a capstone, it is a revision test wearing the name.

The test also knows the difference between working and clicking. B B B B down the page, a forty-question paper finished in four minutes, an ABCD cycle repeated — these are flagged to the class teacher rather than scored as knowledge. Teachers see them. They are not sent home as a verdict on your child.

Per-chapter scheduling

No two chapters get the same schedule.

Chapters here run from 279 to 655 first-pass questions. One fixed schedule strands content on the big ones and produces two-question stubs on the small ones.

Hydrocarbons, searched three ways

4 tests
9 stranded
5 tests
79 stranded
6 tests
29 stranded

Add one test and it gets nine times worse. Add another and it recovers. Each step changes the shape of what is left over, which flips which difficulty runs out first. There is no intuition to hand-tune with — it has to be searched, per chapter, once.

And a bug worth admitting

Question IDs sort as text, so page 10 came before page 3 and page 9 came last. The allocator walked each chapter in the wrong order and never reached the tail.

On Hydrocarbons that left 13 of 30 concepts with zero questions across the entire journey — including two that sit on pages 3 to 9 and are asked most years. Now every concept gets a floor before any concept gets a second helping.

Found by counting, not by noticing. Nobody spotted this by using the product. It surfaced when we counted questions per concept and the totals did not add up — which is the argument for counting things you believe are already fine.

Four things we deleted

Built. Then removed.

Each of these shipped, ran on real student data, and lied. This is the shortest section on the page and the one we would most like you to read.

Predicted NEET rank

We built it. Then we deleted it. One student's own papers produced estimates ranging from rank 30 to rank 21,00,000. A number that unstable tells nobody anything.

A single blended mastery score

A student ten concepts into a sixty-concept chapter, answering at 90%, rendered as roughly 15%. It read as failing. They were doing well.

Accuracy on tiny samples

Two questions is not an accuracy. Below the floor there is no percentage at all now — just a sentence saying why there isn't one yet.

Class rank shown to parents by default

Percentiles motivate the top half and corrode the bottom. It still exists — as a switch the class teacher decides on, not a default we chose for everyone.

These are not coming back. Not as a premium tier, not on request, not with a disclaimer under them. If a number cannot survive being checked, shipping it anyway is just moving the cost onto the student.

The coverage audit

Every paper is checked after the exam.

One question at a time, one thing asked: was this already in NAADI before the exam? A direct hit, a different framing, or a miss.

173 / 180 covered · 149 direct, 24 different framing, 7 missed direct hits alone — 149/180 · 82.8%
Biology
90 of 90
88 direct · 2 different framing · 0 missed
Physics
40 of 45
33 direct · 7 different framing · 5 missed
Chemistry
43 of 45
28 direct · 15 different framing · 2 missed
Direct hit Covered, different framing Missed

All five Physics misses were lab-manual questions. Every NCERT-sourced Physics question on the paper was covered. Lab-manual content is being written now — the honest fix for a gap of that shape. The next paper gets published too, whatever it says.

NEET 2026 Retest · 180 questions · Physics 45, Chemistry 45, Biology 90 · audited 23 June 2026. Coverage counts direct hits plus hits asked from a different framing; the direct-only figure is given above.

You have read the reasons. Now use the thing.

Nothing on this page is a claim you have to accept on trust. Open a chapter, get a question wrong on purpose, and watch what happens three tests later.

No predicted ranks. No guaranteed scores. Not now, not later.