Interview prep checklist

C#, ASP.NET Core, Angular and PostgreSQL topics with a three-state checklist (haan / thoda / naa) and two searchable cheatsheets. Progress is saved in the browser's localStorage.

Plain static site — no framework, no build step.

Run locally

Open index.html in a browser, or serve the folder with any static server:

npx serve .

Editing content

Answers (the main page) are written in markdown under content/answers/, one file per topic, read in filename order. After editing, rebuild:

node tools/build-answers.js

That regenerates js/answers.js — commit both. The markdown format (tables, bullets, code blocks, ? practice prompt, ! interview trap, > memory hook) is described at the top of tools/build-answers.js. The build fails on a question with no answer or a duplicated question.

Give a question a ? ... line when its title is just a term ("CLR") — practice mode shows that line instead, phrased the way an interviewer would ask.

The checklist (DATA) and Cheatsheet 1 (CHEAT) are still plain arrays in js/data.js.

Progress is keyed by the text of each item, so reordering is safe. Editing an item's text (the question, term or topic) starts its tick fresh; editing only an answer keeps it.

Sync between devices

api/progress.js is a Vercel function that stores progress in Redis. No accounts: one device creates a random sync code, other devices enter it. Marks, notes and user records (interview diary entries, own questions, daily practice counts) merge per item by timestamp; a clear or delete wins over an older copy. Record shapes are documented in api/progress.js.

Needs a Redis store connected to the Vercel project (Storage). Either works:

• Redis Cloud (REDIS_URL, a redis:// connection string) — uses the redis package in package.json
• Upstash (KV_REST_API_URL/KV_REST_API_TOKEN or UPSTASH_REDIS_REST_*) — plain REST, preferred if both are set Without it the site still works, and the Sync panel says storage is missing.

Answer judge (practice mode)

Practice answers (typed, or spoken via the browser's speech recognition) are graded by api/judge.js: Claude compares the answer with the card's notes and returns a score, right/partial/wrong, what was covered and missed, mistakes, feedback and a model answer. Mock interview mode asks 8 mixed questions and ends with a hire / next round / not selected verdict from the scores.

Setup — add one key to the Vercel project's environment variables and redeploy:

• GEMINI_API_KEY (free tier from Google AI Studio; Google may use free-tier content to improve its products). Model: GEMINI_MODEL, default gemini-3.8-flash.
• or ANTHROPIC_API_KEY (paid). Model: JUDGE_MODEL, default claude-opus-5. With both set, Claude is used unless JUDGE_PROVIDER=gemini. Limits: JUDGE_PER_HOUR (per IP, default 60) and JUDGE_PER_DAY (site-wide, default 400), kept in the sync Redis. Without a key the page falls back to a rough keyword check and says so; "AI app ke liye copy" puts a grading prompt on the clipboard for any chat app.

PDF of all answers

node tools/print.js out/answers.html
chrome --headless=new --no-pdf-header-footer --user-data-dir=out/profile --print-to-pdf=out/answers.pdf out/answers.html

Every card in full (written order, tables, code, trap, hook) grouped with a clickable index, plus the cheatsheet as an appendix. Diagrams are left out. --user-data-dir keeps it from handing off to an already-open Chrome.

The full PDF is ~9.5 MB — too big for uploads capped near 11 MB once encoded. node tools/print.js out/part1.html 1/3 (then 2/3, 3/3) splits it at group boundaries into ~3 MB parts; numbering continues across parts and the cheatsheet goes in the last one.

Deploy

Hosted on Vercel as a static site: import the repo, framework preset Other, no build command, output directory = repo root.

robots.txt and a noindex meta tag keep it out of search results. Remove both if it should be findable.