CoolFace
Apppublic

vivekananda0010/MY_ASSISTANCE

sourceHugging Faceupdated 2mo agoView on Hugging Face
0likes
App README

PrepApp — Interview Prep Platform

Working, deployable MVP, built entirely on free tiers — $0 to run: auth, DSA (static + AI-generated questions, brute/good/optimal solutions, sandboxed code execution), Aptitude (topic-wise/mixed tests, dual solutions), and a profile/progress page. Build verified clean (npx next build — 0 type errors, 0 compile errors) before handoff.

Free stack: Supabase free tier (Postgres + Auth) · Groq free API (real Llama models, no card required) · Piston public API (code execution, no key needed) · Vercel free tier (hosting). No paid service is required to run this app.

1. Deploy in ~15 minutes

Step 1 — Supabase (database + auth)

  1. 1.Create a project at supabase.com (free tier is enough to start).
  2. 2.Open SQL Editor → paste and run supabase/schema.sql.
  3. 3.Then paste and run supabase/seed.sql (gives you a few real questions on day one).
  4. 4.Authentication → Providers: enable Email, and enable Google if you want Google sign-in (needs a Google OAuth client ID/secret — Supabase's docs walk through this).
  5. 5.Authentication → URL Configuration: add your deployed URL (and http://localhost:3000 for local dev) to Redirect URLs.
  6. 6.Copy from Project Settings → API: Project URL, anon public key, and service_role secret key.

Step 2 — LLM key (free)

  1. 1.Go to console.groq.com/keys, sign up (no card), create an API key.
  2. 2.That's it — Groq's free tier hosts open-weight models at no cost. The app defaults to openai/gpt-oss-120b (Groq's current recommended free workhorse model). Note: Groq deprecated its Llama chat models in June 2026 (fully shut off by August 2026) — if you specifically want Meta Llama weights, check console.groq.com/docs/models for whichever Llama variant is current and set LLM_MODEL to it. Free-tier limits are per-minute/per-day; check current numbers at console.groq.com/settings/limits — the app's own rate limiting (lib/rateLimit.ts) is already tuned to stay well under typical free-tier caps. (Question generation still works without this key — the static seed bank is always available — but AI-generated questions/solutions need it.)

Step 3 — Deploy to Vercel

  1. 1.Push this folder to a GitHub repo.
  2. 2.Import it in vercel.com.
  3. 3.Add environment variables (copy from .env.example):
  4. 4.NEXT_PUBLIC_SUPABASE_URL
  5. 5.NEXT_PUBLIC_SUPABASE_ANON_KEY
  6. 6.SUPABASE_SERVICE_ROLE_KEY
  7. 7.GROQ_API_KEY
  8. 8.Deploy. Done — you have a live URL.

Local development

bash
npm install
cp .env.example .env.local   # fill in real values
npm run dev

2. What's genuinely ready vs. what's next

Ready today:

  • —Email + Google auth, protected routes via middleware
  • —DSA: static bank, AI generation, three-level solutions, code execution against test cases (via the public Piston API), attempt/submission logging
  • —Aptitude: topic-wise/mixed test generation, scoring, per-question correctness
  • —Profile: solved count, accuracy, average aptitude score
  • —Full RLS on every table — every user can only read/write their own rows; questions/solutions are shared read-only content written only by the server
  • —Rate limiting on every LLM-calling endpoint — per-user daily caps (20 question generations, 30 solution generations, 5 AI-assisted test builds), enforced atomically in Postgres so it can't be raced. Tune the numbers in lib/rateLimit.ts once you see real usage/cost.

Deliberately deferred (flagged, not hidden):

  • —System Design / SQL / NoSQL / ML / DL / CV / RAG — DB schema already supports these topics (topic enum includes them), UI shows them as "coming soon." Wiring up content generation for each is the same pattern as DSA — copy app/dsa and the two question/solution API routes, change the topic value and prompt.
  • —Production code sandbox — currently uses the free public Piston API (rate-limited, fine for early users/testing). Before real traffic, self-host Piston (one docker run, see their repo) and point PISTON_URL at it.
  • —RAG / vector DB — not wired up. Needed for the "conceptual Q&A" module; add pgvector to Supabase (it's a one-click extension) when you build that module.
  • —AI code review (beyond pass/fail) — the schema has an ai_feedback column on submissions ready for it; not called yet.
  • —Scaling past Groq's free tier — fine for an MVP/early users. If you outgrow it, either move to Groq's paid tier or point lib/llm.ts at a self-hosted Llama (Ollama/vLLM) — no other code changes needed, same generateJSON interface.

3. Security notes

  • —SUPABASE_SERVICE_ROLE_KEY bypasses Row Level Security — it's used only in server-side API routes, never sent to the browser. Keep it out of any NEXT_PUBLIC_* variable.
  • —User code submissions never execute on your server — they're forwarded to the Piston sandbox API. Your server only sends/receives text.
  • —All protected pages check auth server-side (middleware + per-page checks), so there's no client-only gate that could be bypassed by disabling JS.

4. Tech stack

Next.js 14 (App Router, TypeScript) · Supabase (Postgres + Auth, free tier) · Groq API (question/solution generation, free tier, open-weight models) · Piston API (code execution, free, no key) · Tailwind CSS · Vercel (hosting, free tier)