Most learning systems know the subject. We know the learner.
Trangy Labs builds the technology that understands how individual learners think. Their specific misconceptions, the exact moments their reasoning breaks, what kind of intervention actually works for them. The signal that every learning system in India has been missing.
Meet Sahiprep.
Sahiprep is the first product built on Trangy Labs' learner modelling engine, designed for competitive exam aspirants across India.
Learners are not categories. They are patterns.
A short, honest account of what we mean by learner modelling, what it is, what it is not, and why the distinction is the entire game.
The system has been blaming the wrong person.
Every learning platform in India operates on the same broken assumption. One curriculum, delivered the same way, will produce good outcomes for everyone. Students who do not fit that curriculum struggle, drop out, or compensate with brute-force study hours.
The system never adapts. The learner is expected to. And when the learner fails, the system blames the learner.
Recommendation is not personalisation. Modelling is.
Most platforms mean something simple by personalisation. Finished one video, here is the next. Got this question wrong, here is a similar one. That is content sequencing. True personalisation requires the system to know how a specific person thinks.
What others do
Track topic-level scores. Recommend the next piece of content. Repeat.
What we do
Build a multidimensional model of how a person thinks, where they get stuck, and what would unstick them.
None of this is captured by existing platforms. They report scores. We model behaviour. Here is what that looks like in practice:
This learner understands the concept but consistently makes calculation errors under time pressure.
This learner falls for engineered traps in a specific class of questions across multiple topics.
This learner skips time-intensive problems even when shortcuts would have made them tractable.
This learner's grammar errors cluster around a pattern invisible at the topic level.
Twenty-one ways of looking at a question.
Most platforms tag content across two dimensions: topic and difficulty. A question is "Algebra, Hard," and that is the end of what the system knows about it.
Our cognitive tagging system classifies content across 21 structural and behavioural dimensions. How a question tests reasoning, what kind of trap it sets, what shortcut it rewards, what time-pressure dynamic it triggers, what misconception it surfaces.
We are Gen-AI native by design, built around frontier large language models. Every learner makes the model smarter.
Five things we keep non-negotiable.
Technology that compounds
Every additional learner makes the system more accurate for every other. The asset is not content. It is the model.
Measure what matters
Outcomes, retention, behavioural changes our diagnostics drive. Not time-on-app or vanity metrics dressed up as engagement.
Charge sustainably
Bootstrapped and committed to a revenue-generating business from day one. Not a venture-funded scale experiment.
Share what we learn
Our methodology, taxonomy choices, and diagnostic approach are documented openly. The field needs rigorous thinking, not secrecy.
Respect the learner
No trivial gamification. No engagement manipulation. Tools we would want our own siblings using.
Write to us. We read everything.
We are a small team. No contact form, no chatbot, no auto-reply. Just an inbox we genuinely watch.
Partnerships
Schools, coaching bodies, certification providers. Tell us what you are solving.
Press and ecosystem
Writing about Indian EdTech or learner modelling. We are happy to talk on record.
Everything else
Questions, disagreements, a sharp essay. Send it. We respond personally.