PJAM is a mobile-first game that teaches complete beginners the real skill of AI prompting. No multiple choice, no theory. You write ONE prompt, our AI judge scores it out of 100 (Accuracy 60 · Efficiency 25 · Style 15), and you either pass the level or go again. One shot. Just like real life.
For the Hacktoberfest/Día de los Muertos season we shipped a limited-time event layer: The Ofrenda Track. The app wears its purple identity with marigold gold, candles and papel-picado warmth, a seasonal celebration of honoring the past, auto-scheduled to end Nov 2 via a single event flag in our theme config.
🛠 What's under the hood
- Next.js 14 App Router + Tailwind, Supabase backend, deployed on Vercel
- AI judging pipeline: Gemini first, OpenAI fallback, offline heuristic last, so the game never breaks, even mid-tournament
- lib/theme.js is the single source of truth for game numbers, themes and judge instructions
📡 Why Sentry matters to us
PJAM is literally a game about talking to an AI, so we trace every single challenge run end-to-end with Sentry agent tracing. Each run becomes one trace: prompt → AI call → grading, with the challenge id, player level, estimated tokens, latency and final score as span attributes. That means we can see exactly which levels the AI judges struggle with, where latency spikes, and how players' prompts evolve, real product insight, not just crash reports. (We never log players' raw prompts, only metrics. Privacy first.)
🔗 Repo: 🔧 [https://github.com/Collinxweb/PJAM]
🌐 Live:[https://www.pjam.site/]
👻 what's live: Ghost Interactions levels
The Ofrenda Track's next act: levels where you prompt to summon, question and calm the spirits, guided "ghost interactions" that teach constraint-based prompting (tone control, role play, memory of previous prompts) through the Día de los Muertos theme. You'll interview the ancestor, get graded on how precisely you prompt. Honor the past, one prompt at a time. 🕯
Built with ❤️ by Collins Alfred, builder & security researcher.
Top comments (0)