Skip to content
AI

Prompt Injection Playground

Attack your own LLM app with known injection patterns and see which ones get through.

Intermediate18h1–3 people16/20 fit
TL;DR

Build Prompt Injection Playground for developers building llm apps in ~18h with 1–3 people. Stack: LLM + Node.js + Next.js.

Why it wins: Security plus AI is a hot track and the attack demo is dramatic.

Build flow

01 · Problem
Teams ship chatbots without testing whether a web page or email can hijack them.
02 · Solution
Run a library of injection payloads against a configurable system prompt and tools, score which attacks succeed and suggest mitigations.
03 · Build
  1. 1Collect public injection patterns
  2. 2Run each against a target prompt
  3. 3Detect leaks with canary tokens
  4. 4Show a report with fixes
04 · Demo
Show a calendar-assistant bot leaking its secret before and after a mitigation.

MVP checklist

  • Payload library
  • Attack success score
  • Mitigation tips
  • Report export

Tech stack

TechRole
LLMLanguage model
Node.jsBackend runtime
Next.jsFrontend + API routes

Related ideas

More AI

  • AI

    Blind Model Arena for Your Prompt

    Compare several models' answers to your own prompt without seeing which is which

    Intermediate12h1–2
    Vercel AI SDKNext.jsSQLite
    17/20

  • AI

    Ask Your CSV in Plain English

    Upload a CSV, ask a question and get a chart plus the SQL that produced it

    Intermediate14h1–2
    DuckDBLLMCharts
    18/20

  • AI

    Document Q&A

    Lets people ask questions about long documents instantly

    Intermediate24h1–3
    Next.jsRAGLLM
    16/20