Chapter accuracy
Per-chapter and per-unit correctness across every attempt, ranked worst-first.
JEEnify and NEETnify look like two companies’ work. They are one system with two syllabi bolted on — which is why a fix to the exam player, the analytics model or the delivery layer improves both the same week.
The hard part of an exam platform is not the app. It is turning decades of papers, textbooks and coaching material into structured items a machine can reason about and a student can trust.
A past paper, a practice sheet or a centre’s own PDF lands in the queue.
Text, figures and inline images are pulled out and each question is reconstructed as structured JSON.
Worked, stepwise solutions are generated against a fixed format standard and validated for shape.
A reviewer approves or corrects, then the item goes live under the right subject and syllabus unit.
Students lose marks to an unfamiliar interface. We removed that variable by reproducing the portal rather than approximating it — the same palette states, the same navigation, the same unforgiving clock.
A total score tells a student almost nothing. What changes behaviour is knowing which chapter is bleeding marks, how long they are over-spending on it, and what the trend looks like against the exam that is coming.
Per-chapter and per-unit correctness across every attempt, ranked worst-first.
Time-per-question against the budget the real paper allows, flagged where it exceeds it.
Marks-vs-percentile mapping built from historical trends, applied to the student’s current score.
A single honest number combining coverage, accuracy, pace and recency — designed to be uncomfortable early.
A centre is a first-class tenant, not a folder. It has its own URL, its own branding, its own admin console, its own members, its own papers and its own analytics — and no visibility into any other centre’s data.
Centre-authored content never joins the shared bank unless the centre explicitly asks us to publish it.
Most of our students are not on office Wi-Fi. The platform is engineered around that constraint rather than apologising for it.
Add to home screen on Android, iOS or desktop. Service-worker caching keeps the shell and the current paper available through a dropout.
Aggressive payload budgets, image optimisation and web-vitals monitoring in production — measured on real devices, not a laptop.
Containerised services on managed cloud with a managed PostgreSQL primary, automated backups and health checks — sized to absorb a live-mock spike.
| Layer | What we run | Why |
|---|---|---|
| Application | Next.js (App Router) with React and TypeScript in strict mode | Server rendering for crawlability and cold-start speed; one language across the stack |
| Content services | Python 3.13 extraction and solution pipeline | The document, OCR and ML tooling lives in Python |
| Model layer | Frontier vision-capable LLMs behind our own prompt and validation standard | Figures matter in Physics and Biology; a text-only model is not enough |
| Data | Managed PostgreSQL, single writer, automated backups | Attempts and results are financial-grade records; we treat them that way |
| Delivery | Containers on managed cloud, CDN-fronted static assets, PWA service worker | Predictable scaling through live-mock peaks |
We hold minors’ data and we publish numbers students make decisions on. Both deserve more care than a marketing page usually admits.
Found something that looks wrong? Write to security@skelro.com. We read every report and we do not pursue researchers who disclose in good faith.
Skelro Centers gives a coaching institute the whole platform — question bank, paper builder, exam engine and analytics — on its own URL.