Full-Stack Developer · Saaradaa Learknowations (SLPL) — Hyderabad, India
B.Tech in Mechanical Engineering — not computer science. The production stack below is self-taught, learned by building and shipping a live multi-tenant exam platform, essentially solo, over nine months. The company defined the vision; I owned the execution — architecture, code, and deployment.
ScrollOne platform serving multiple schools: a shared global question repository, live timed exams with per-click autosave, role-separated administration, and analytics for both students and institutions — later extended into a public B2C practice platform with live payments, on the same engine. I went in never having deployed anything to production. I came out owning all of it.
First production deploy failed in an environment-config cascade — missing DB connection string, wrong S3 credentials, each error masking the next. Built a strict environment checklist that became the platform's permanent deployment runbook.
MySQL connection pool exhaustion under concurrent student submissions — cascading request failures mid-exam. Diagnosed the root cause, introduced gevent async workers with a dynamically managed pool; later moved production serving to waitress as the platform matured.
Per-school data isolation enforced at the route layer — every tenant-scoped route checks
school_id before the query runs. That's route-level discipline, not schema-level
enforcement; pushing the check into the model layer is the named next hardening step.
A shared question corrected once propagates to every school's exam at query time — no fan-out writes. Schools can still override marks locally: the override is scoped to one school's exam and never touches the master.
Click-triggered saves send only the single answer that changed — not the full exam state. Third iteration of the design, after timer-based polling and full-array saves. Enables Lazy Finalization: auto-scoring from the last recorded click-state when a connection drops.
Student IDs embedded class level as a static string — a promoted student would keep a stale ID forever. Caught after real students had already been assigned IDs; migrated existing records to a class-agnostic scheme (school code + scoped sequence).
Schema validation before any write, duplicate detection via a normalized composite key (text / class / subject / chapter / options), and row-level error reporting — good rows insert, bad rows come back with reasons. No silent failures, no total rollback.
Institution-wide view added recently, distinct from the earlier student-facing analytics: mean, median and distribution stats, subject heatmaps, question-level miss rates — computed once at result-release and cached server-side.
Self-registration, exam-field selection (NEET, JEE, UPSC, banking), and live Razorpay payments — order creation, signature verification, subscriptions — reusing the same question repository and exam engine underneath. No rebuild.
Self-taught · shipped to real users
Trained at NIIT
The split is deliberate. The left column was learned under real pressure — production failures, live users, no senior engineer to hand things to. The right column is formal training: a Java full-stack course at NIIT, applied in team and solo course projects.
Web app for notes and task reminders. Worked across frontend, backend, and database: user authentication, note creation, editing, archiving and deletion, with filtering by date and priority.
Product catalogue with multiple views and an ordering flow with distinct forms for users and owners. Role-based: users place orders; owners manage and review them under separate admin authorization.
Also published — Saaradaa Journal of Interdisciplinary Studies
Online Exam Portal to Manage Monthly Exams and Assignments Between Different Schools Vol 1, Issue 2 · pp. 25–36 · the SL-Exams platform, written up as a paper ↗ The Burden of Benevolence: Expectancy Traps, Pathological Altruism, and the Overthinker's Dilemma Vol 2, Issue 1 · pp. 90–97 ↗ORCID: 0009-0006-9997-5183