PitchPrep (beta): DECA-style practice
Cluster exams weighted to the official 2026–27 blueprint, spaced-repetition vocab, and live voice judges you present to out loud: role-plays for every DECA role-play event, plus the Pitch Room (DECA Judge for business plans, Q&A Drill, Shark Tank, Cold Call, Professional Selling). See PLAN.md for the roadmap.
Run it
npm install
npm run dev:vinext # http://localhost:3001 (needs `npx wrangler login` for the AI binding)
npm run deploy:vinext # deploys to Cloudflare Workers
No API key needed. The AI runs on Cloudflare Workers AI (@cf/openai/gpt-oss-120b for the judge and scoring, MeloTTS for the judge's voice) on the account's free daily allowance. app/api/ai/route.ts stops at 9,000 of the 10,000 free neurons a day, so it never bills; a full role-play costs about 360, so roughly 25 sessions a day. The limit resets at 8 PM Eastern.
What's in the beta
| Area | Where |
|---|
| Event / level picker, weak-area dashboard | app/page.tsx |
| Blueprint mock exams, topic drills, missed-question review | app/exam/page.tsx, lib/deca.ts |
| Flashcards (Leitner boxes) and 60s Blitz | app/vocab/page.tsx |
| Role-play: scenario from real PIs → prep timer → live voice judge with 2 follow-ups → 2026–27 evaluation form | app/roleplay/page.tsx |
| Pitch Room: DECA Judge, Q&A Drill, Shark Tank, Cold Call, Professional Selling | app/room/page.tsx, lib/personas.ts |
| Live voice engine (mic, speech-to-text, turn-taking, recording, delivery metrics) | lib/room.ts, components/LiveRoom.tsx |
| Judge prompts and scoring | lib/judge.ts, app/api/ai/route.ts |
| Official 2026–27 evaluation forms | lib/forms.ts (npx tsx lib/forms.ts checks every form totals 100) |
| Question bank (818), vocab, blueprint, BA Core PIs | data/ |
Members sign in with email and password. Only emails on the member list (the Members page, admins only) can get in: adding someone gives a one-time invite link where they set their password, and New link resets a forgotten one. Each member's history and progress are saved to their own account in the pitchprep D1 database (schema.sql), so nobody sees anyone else's.
Growing the question bank
data/items.generated.json holds 401 original items added Oct 2026 so mock exams don't repeat. Have a strong competitor review them, then move the good ones into data/items.json. (scripts/generate-items.mjs still uses the Anthropic API and needs a key.)
Content rule: everything must be original. Never paste in DECA+ exams/scenarios or questions from other prep sites.
Launch checklist (what's left for a human)
- Members. Add each member on the Members page and send them their invite link.
- Officer test run (about 30 minutes). Have 2–3 strong competitors each run 2–3 sessions across events and levels. Check that scenarios read like real DECA role-plays, the judge's questions respond to what was said, and scores land within roughly ±10 of an experienced judge.
- Advisor OK. Show your advisor the Privacy page (
/privacy). Answers are sent to an AI service, members are minors, and each member's results are stored in their account.
- Share the URL with the members you added. Voice works best in Chrome or Edge on a laptop; on phones, typing is more reliable.
Re-check the scoring forms each season. lib/forms.ts follows the 2026–27 DECA Guide and DECA's 2026 sample events: Novice/Developing/Proficient/Exemplary levels, Solution and Career Competencies rows, Team Decision Making cut to 5 PIs. Marketing Management Series was dropped for 2026–27.
Checking the data
npm run check validates every question and prints how full each cluster's mock exam is. Run it after adding questions.
