Every question. Reviewed before it reaches you.
QTG draws every question live from qSarthi Kota's question bank — tagged on 25+ details, in three languages, and released only after about 11 AI reviews and 4 human expert sign-offs.
The road every question takes
Different AI models at every stage, and a human expert who has to sign off before it moves on.
-
DTP verify
2–3 AI reviewsLayout, figures and maths checked against the source
+ human sign-off -
Review L1
5 AI reviewsCorrectness, answer key and classification — five models vote
+ human sign-off -
Review L2
3 AI reviewsDeep expert pass on language, difficulty and solution
+ human sign-off -
Final review
1 AI reviewA last full-question audit before release
+ human sign-off - ✓RELEASED
One question, written once, used everywhere
Every question lives once in the qSarthi question bank (QBMS) — written, versioned and classified, in English, Hindi and Gujarati. Your papers draw on it live, so a correction reaches every draft.
🌐Three languages
- English, Hindi and Gujarati versions of the same question, option by option.
- Each language is reviewed on its own before it can be used.
- Print in one language, or English with Hindi or Gujarati side by side.
🧪Science that prints right
- Maths typeset exactly as written (KaTeX).
- Chemistry structures drawn from their formula, the same in every language.
- Figures keep their size; correct Hindi conjuncts in print.
🧩Many question types
- Single-correct MCQ, multiple-correct, numerical / integer.
- Assertion–reason, statement type, matrix match.
- Fill in the blank, true / false, case based and more.
🎯Exams covered
- NEET, JEE Main, JEE Advanced, CUET, BITSAT, KCET, MHT CET, GUJCET, WBJEE.
- CBSE and ICSE boards, Classes 6–12.
- Olympiads and Foundation: NSEJS, NSTSE, IMO, NSO and more.
📜Previous-year questions
- PYQs carry their exam and year — "NEET 2021".
- Filter a paper to PYQs only, or by year.
- The PYQ tag can be printed beside the question.
🔁Always up to date
- Every question has a version history.
- A corrected question refreshes into your draft papers with Sync.
- Finalized papers never change underneath you.
25+ details on every question
Rich tagging is what lets you ask for "10 hard, high-yield application questions from Rotational Motion". Tagging is checked at review — an under-classified question can't be released.
| Detail | Values |
|---|---|
| Difficulty | Easy · Medium · Hard |
| Importance | High yield · Medium yield · Low yield |
| Bloom's level | Remember · Understand · Apply · Analyse · Evaluate · Create |
| Key skill | Conceptual understanding · Problem solving · Logical reasoning · Analytical thinking · Formula application · Data interpretation · Memory & recall |
| Time needed | Quick (under 45 s) · Standard (about a minute) · Lengthy (2–4 min) |
| Syllabus | Class → subject → chapter → topic → sub-topic → micro-concept |
About 11 AI reviews and 4 human sign-offs
A question moves through four stages before it is released. At each stage AI models check it first, then a subject expert signs it off. AI only suggests — a person approves every change.
1DTP verify
2–3 AI reviews compare the layout, figures and maths with the source page, using vision models. Then a person checks it.
2Level-1 review
AI checks correctness, the answer key and the classification. A subject expert reviews the question and approves it.
3Level-2 · Super-expert audit
A sceptical second pass on the answer, options, solution, figures, classification and every language. A low score needs an expert's written reason to move on.
4Release audit
A last full check that separates release-blocking defects from cosmetic notes and ends in release or hold. Only then does the question reach papers.
Hard gates
A likely duplicate, missing classification or a failed answer-key check stops a question from moving forward. A low AI score needs a written reason, or a fresh audit if the question was edited.
Humans commit every change
AI suggestions always arrive as pending. A subject expert reads them and applies, edits or rejects each one. Teachers using QTG can report a question at any time, and it comes straight back into review.
A fleet of models — so no single blind spot survives
The question bank runs 13 AI models from 9 makers. Each question is checked by several different models, so a mistake one model misses is caught by another.
GPT-6 Luna Pro
OpenAI. Checks the answer, options, solution and classification of a question.
Claude Haiku 5.5 · Gemini 3.8 Flash
Anthropic and Google. Fast second and third opinions in review.
DeepSeek · GLM · MiniMax · Qwen
Four more makers, each trained differently — more independent eyes on every question.
Open models: Gemma 4 · Nemotron
Google's Gemma 4 and NVIDIA's Nemotron 3 Ultra and 3.5 Super add further checks at large volume.
Nano Banana 2 · Meta Muse
Google and Meta image models behind AI diagrams and figure editing.
The models in use
| Model | Maker | Used for |
|---|---|---|
| GPT-6 Luna Pro | OpenAI | Question review |
| Claude Haiku 5.5 | Anthropic | Question review |
| Gemini 3.8 Flash | Question review | |
| DeepSeek V4.1 | DeepSeek | Question review |
| GLM-5.3 | Z.ai | Question review |
| MiniMax M3.1 | MiniMax | Question review |
| Qwen 3.8 | Alibaba | Question review |
| Gemma 4 31B (open) | Question review | |
| Gemma 4 26B (open) | Question review | |
| Nemotron 3 Ultra (open) | NVIDIA | Question review |
| Nemotron 3.5 Super (open) | NVIDIA | Question review |
| Nano Banana 2 | Diagrams and images | |
| Meta Muse | Meta | Diagrams and images |
Models are chosen per review and swapped as better ones arrive.
What QBMS is built for
QBMS is the system behind the bank, built in-house by our own team.
10 crore questions
Designed to hold up to 10 crore (100 million) questions.
Deep AI integration
Any AI model can be plugged in and used across the bank.
API-first
Fully API-based, so it connects with any existing or new app.
Multi-tenant
Every organisation's data is kept separate from everyone else's.
Cloud native
Built on the latest tech stack, and deployable anywhere.
Secure by design
Data security is taken care of end to end.