CalcSprint — Banking Calculation Trainer

A responsive, gamified calculation-practice frontend for SBI Clerk, IBPS Clerk/CSA and IBPS RRB Office Assistant preparation.

Included in this MVP

• Tables 1–30
• Squares 1–40
• Cubes 1–20
• Fraction → percentage
• Percentage → fraction
• Powers of 2 (2¹–2¹⁰)
• Perfect square roots up to 2500
• Addition, subtraction, 2-digit multiplication and exact division
• Mixed practice and adaptive difficulty
• 1 / 2 / 5 / 10 minute speed challenges
• Banking calculation drill for DI, percentages, ratios and simplification
• Keyboard-first practice (Enter submit, Esc pause, Space resume)
• Correct/incorrect feedback and optional sound
• Streaks, XP and levels
• Daily 50-question challenge with a 500 XP completion bonus
• Mistake review and retry
• Weak-area weighted practice
• Personal bests and achievements
• Performance analytics with Recharts
• Accounts (sign up, log in, log out); progress and mock tests are saved per user
• Responsive sidebar/mobile navigation
• Every screen has its own URL: sessions keep their settings in the query (/practice/session?cat=tables&diff=medium&n=20) and results live at <setup>/result/[sessionId]
• Mock tests from PDFs (Quant, Reasoning, English) in an IBPS / SBI-style exam interface — see below

Stack

• Next.js App Router
• React
• TypeScript
• Tailwind CSS
• shadcn-compatible local UI primitives (src/components/ui + components.json)
• Lucide React
• Recharts

Data is stored in Upstash Redis (or in a local .data/ folder during development). Mock tests also need a Gemini API key.

Run locally

npm install
npm run dev

Then open:

http://localhost:3000

Mock tests

1. Put a Gemini key in .env.local (see .env.example):

   GEMINI_API_KEY=your-key
   

2. Run npm run dev, log in, then open Mock Tests, upload a prelims PDF, then generate as many tests as you like.

How it works:

• The PDF is sent to Gemini once (src/app/api/mock/sources/route.ts). Every generated test follows the paper's exact pattern: the same sections, topics, question counts and order (paperPattern / buildMockTest in src/lib/mock-generator.ts).
• Calculation questions (simplification and arithmetic) become templates: variables with the original numbers, derived values for questions with the ? in the middle, and an answer formula evaluated safely by src/lib/expression.ts. A template is used only if feeding it the original numbers gives the original answer. New numbers keep the original's digit count and roundness, the answer stays the same size, and wrong options sit as close to the answer as the original's did.
• Everything else (DI, Reasoning, English) gets AI-written versions, written after upload from the paper page a few sets per request. A version is kept only if it is within ~20% of the original's passage and question length and a separate review pass, solving it without the answer key, agrees with the key and rates it no easier and no simpler in wording than the original.
• Papers, tests and attempts are saved to the user's account. Deleting a paper also deletes its tests and attempts. PDFs can be up to 25 MB. A paper is read in parts (an outline call, then about 15 questions per call, each with the whole PDF attached through the Gemini Files API) so long papers never overflow one answer; progress is saved after every part. With several keys in GEMINI_API_KEYS, a key that runs out of quota or is revoked is rested and the work continues on the next key (src/lib/server/gemini-keys.ts). Vercel functions only accept 4.5 MB per request, so on Vercel the browser uploads the PDF to Vercel Blob first; the server reads it from there and deletes it. Connect a Blob store (Storage → Blob, choose Private) to the project for PDFs over 4 MB; both newer stores (BLOB_STORE_ID, signed in through Vercel OIDC) and older ones (BLOB_READ_WRITE_TOKEN) work.
• The exam screen copies the IBPS / SBI interface: section tabs, sectional or composite timing, question palette with the official status legend, Save & Next, Mark for Review & Next, Clear Response and the pre-submit summary. Marking is +1 / −0.25. Progress autosaves, so a reload resumes the same question.

Production validation

npm run lint
npx tsc --noEmit
npm run build

Accounts and data storage

• Every page except the landing page, /login and /signup needs an account. src/proxy.ts redirects visitors without a session cookie; every API route checks the session itself (src/lib/server/auth.ts).
• Passwords are hashed with scrypt. Sessions are random tokens in an httpOnly cookie, stored hashed for 30 days. Logins are limited to 10 wrong attempts per 15 minutes per email.
• All data lives under per-user keys (src/lib/server/store.ts), so one account can never read another's.
• Practice sessions are sent to the server as events and applied there (applySession in src/lib/stats.ts), so two devices don't overwrite each other.
• Progress saved in the browser before accounts existed is imported into the account on first login, if the account has no progress yet.
• Settings → Storage shows what the account uses and has delete buttons for practice history, mock results, unstarted tests and each paper.

Database

• Vercel: in the project, open Storage, connect Upstash for Redis and redeploy. It sets KV_REST_API_URL and KV_REST_API_TOKEN.
• Local: without those variables, data is kept in .data/ (git-ignored). To use the Vercel database locally instead, run vercel env pull .env.local.

Main source areas

src/
  app/                    # Next.js routes
  components/
    app-shell.tsx         # Responsive application shell
    exam/                 # IBPS / SBI-style mock test interface
    session-engine.tsx    # Core practice/game loop
    session-route.tsx     # Builds a session from its URL
    session-setup.tsx     # Practice configuration
    ui/                   # Local UI primitives
  constants/              # Category definitions
  context/                # Auth and stats providers (synced with the server)
  lib/
    question-generator.ts # Dynamic question generation + answer checking
    session-url.ts        # Session settings <-> URL query
    gemini.ts             # Server-only Gemini calls for mock tests
    server/               # Auth, database access and API helpers (server only)
    mock-generator.ts     # Builds mock tests from parsed papers
    stats.ts              # Analytics, XP, levels, weak areas
  types/                  # Shared TypeScript models

Notes

• Question difficulty adapts during a session based on recent accuracy and response time.
• Weak Area Training weights categories with low accuracy / slow average response time.
• Pausing hides the current question and freezes its timer.
• Fraction answers accept equivalent simplified or unsimplified fractions because checking uses cross multiplication.