Prompt Finder

Challenge

Managing hundreds of ChatGPT prompts across multiple PDFs is a nightmare. Each prompt is buried in a document, categories are mixed together, and finding the right prompt for a task means scrolling through pages of unrelated content. The challenge was to extract, categorise, and surface all prompts in an instantly searchable interface that works offline with zero setup — no server, no build step, no dependencies.

Approach

I wrote a Python script to parse the source PDFs, extract each prompt with its metadata, and generate a self-contained HTML file with all data inlined as JSON. The result is two static HTML files that open in any browser. The main view features a search bar that filters prompts in real-time, category chips for browsing, and a results list with source attribution. The gallery view adds a grid layout with bookmarking (saved to localStorage) and a filtered category page system. Both include keyboard navigation, smooth scroll animations, and a dark theme that respects system preferences. The entire app loads instantly — no network requests, no loading spinners. For distribution, I generated a zip containing just the two HTML files — 462 KB compressed. Unzip and open in any browser.

Key Code

// Core search filter — runs on every keystroke
const q = query.value.toLowerCase().trim();
const filtered = DATA.filter(p => {
  if (q === '') return !cat || p.category === cat;
  const catOk = !cat || p.category === cat;
  const txtOk = p.text.toLowerCase().includes(q) ||
    p.keywords.some(k => k.includes(q));
  return catOk && txtOk;
});

Results

• 3,273 prompts extracted from PDFs into a searchable interface
• 7 categories with colour-coded filter chips for browsing
• Instant real-time search across prompt text and keywords
• Bookmarking system with localStorage persistence
• Zero dependencies — works offline, no server or build required
• 462 KB zip distribution for instant deployment

Key Learnings

• Static HTML with inlined JSON is the simplest possible deployment — no infra to maintain
• For content-heavy tools, a search bar + category filter covers 90% of user needs
• localStorage is perfect for lightweight personalisation (bookmarks) with zero backend
• Data curation matters as much as the interface — clean categories make or break usability

Gallery