01
Grounded, not guessed
Every answer traces back to source material: 25+ years of past papers, NCERT text, official marking schemes. If a claim can't be grounded, it doesn't ship.
Intelligence is grey.
A black box hands you an answer. A white paper hands you a theory. Grey matter is where reasoning actually happens: layered, patient, and willing to show its work.
That middle part is what we build. Grey Neural AI is a small applied AI company in India making systems for people who can't afford a wrong answer. Today that means students sitting the hardest exams in the country. Tomorrow it means anywhere a good explanation changes an outcome.
Every answer traces back to source material: 25+ years of past papers, NCERT text, official marking schemes. If a claim can't be grounded, it doesn't ship.
The model explains until it lands. At 2am. For the fifth time. From a different angle if the first four didn't work, and without a trace of impatience.
We publish our accuracy, including the subjects where we're weakest. Our Mathematics coverage sits at 69% against 92% for Physics. We'd rather you heard that from us.
Three layers do the work. Each one is replaceable, measurable, and answerable for its part of the result.
Different questions want different minds. A vector-geometry proof and an organic mechanism don't belong on the same model. We route per question, not per product, and we re-route when a model stops being the best answer.
Agents trained on 25+ years of JEE and NEET papers: the phrasing, the traps, the marking scheme, the way an examiner actually reads a step. Generic models know the physics. These know the exam.
Every step is anchored to source and rendered as working, not verdict. A student can follow it, a teacher can audit it, and we can measure it. That last part is why the layer exists.
Before JEE Main 2026, our engine generated a set of prediction papers. Afterwards we ran every real question against them. Here is what came back.
Measured by L2 distance over 768-dimension sentence embeddings, the same similarity engine that runs inside the product. To be precise about it: we predict concepts, not questions. There were no exact duplicates, and Mathematics coverage came in at 69.4% against 92.4% for Physics. Full methodology available on request.
Each one is shaped around how that exam is actually sat, not around what was easiest for us to build twice.
The full preparation stack: ask anything, practise everything, and sit as many mocks as you have evenings for.
Rebuilt for a paper that rewards recall as much as reasoning, so the answers stay anchored to the book the exam is written from.
Every JEE past question, chapter-wise, in your pocket. The narrow tool for the students who just want the papers done properly.
It's 11pm. The doubt is small, the exam isn't, and there's nobody awake to ask.
Most students don't fail on ability. They stall on a single step nobody was around to unblock, and lose the week that followed.
The stack isn't exam-shaped. It points at any field where a good explanation changes an outcome.
Real syllabi are messy, inconsistent and full of near-duplicates. We work on retrieval and evaluation harnesses that catch confident nonsense before a student ever sees it.
We take on a small number of new projects each year, for teams with a problem worth solving properly and the patience to solve it properly.
Healthcare guidance, legal comprehension, technical training: anywhere the gap between an answer and an understood answer is where the harm lives.
Grey Neural AI is an applied AI company based in India. It builds systems that explain their reasoning rather than only returning answers. Its flagship product family, JEEnius, provides AI tutoring, prediction papers and mock tests for students preparing for the JEE and NEET entrance exams.
No. Grey Neural AI, sometimes written GreyNeural, is an unrelated company. It is an applied AI company based in India, and it builds the JEEnius suite for JEE and NEET preparation. It has no connection to GreyNeurons Consulting, Grayscale AI, GreyLabs AI or Grey AI. The official site is greyneural.ai.
Measured against JEE Main 2026, 83.1% of the 728 actual questions fell within topic areas JEEnius predicted, including 92.4% of Physics questions. Sixty questions matched a predicted question at over 80% semantic similarity. Mathematics was weakest at 69.4%. These are topic-coverage figures, not exact-question matches.
A general chatbot answers from broad training data. JEEnius routes each question to a model chosen for that subject, then answers through agents fine-tuned on 25 years of JEE and NEET papers, anchored to NCERT text and official marking schemes. Every step is shown as working, not verdict.
AI can predict topics reliably and exact questions rarely. Against JEE Main 2026, JEEnius predicted the topic areas of 83.1% of the 728 actual questions, but produced no exact duplicates. Sixty questions matched a predicted question above 80% semantic similarity. Treat prediction as a way to prioritise revision, not to skip syllabus.
2,279,743 candidates registered for NEET UG 2026 and 1,355,293 registered for JEE Main 2026 Session 1, according to India's National Testing Agency. Together these are among the largest annual competitive examination cohorts in the world, at more than three and a half million registrations a year.
JEEnius is free to start on both web and Android. The JEE, NEET and PYQ Android apps and both web apps are available now. Students can practise past papers, ask doubts and sit mock tests without paying upfront.
Yes. Alongside its own products, Grey Neural AI takes on a small number of partner builds each year, for teams with a problem worth solving properly. Work is scoped on the same principle as the products: answers that can be checked, not answers that have to be trusted.
We take on a small number of new projects each year, alongside our own products. Tell us what's breaking. If we're the right team for it we'll say so, and if we're not, we'll tell you that too.