AI-powered voice roleplay
Practice real customer conversations, out loud.
Live speech-to-text feeds a stateful, in-character AI customer whose reply is spoken back to you. Time-boxed 5-minute sessions in Bahasa Indonesia.
Barista training, the fun way
Cangkruk turns a café's own SOP & menu into gamified, level-based training — then becomes the difficult customer you practice against, and grades you.
Busy managers have little time for hands-on coaching, so new baristas memorize SOPs or learn on real customers — inconsistent service, avoidable mistakes, low confidence. Cangkruk gives them a safe place to practice, and gives managers their time back.
Upload your docs, auto-generate the training, simulate the real job, get graded.
A manager uploads the café's real PDF/DOCX SOP + menu.
Clean text is pulled from those files on-device — with OCR fallback for scans.
An LLM restructures the raw text into clean, leveled learning modules.
Trainees read the generated material, organized into progressive levels.
Trainees roleplay out loud against escalating AI customer personas.
The AI grades the transcript with a summary + coaching feedback.
A home-screen widget + notifications nudge daily practice.
Try it — tap through, or talk to the customer
Untuk pelanggan yang tidak minum kopi, kita punya pilihan terbaik:
Pelanggan Ramah
Pelanggan santai yang baru pertama kali datang. Penasaran dan suka tanya rekomendasi.
00:00
TEKAN UNTUK BERBICARA
Durasi latihan
03:17
Kamu membuka percakapan dengan sapaan yang hangat dan langsung menawarkan menu yang memang ada, bukan jawaban mengambang. Waktu pelanggan bilang tidak minum kopi, kamu berpindah ke daftar non-kopi alih-alih memaksakan kopi — bacaan situasi yang bagus. Mengulang pesanan di akhir juga penutup yang tepat.
Sebutkan karakter rasanya, bukan cuma nama minumannya — “earthy dan agak manis” lebih membantu pelanggan daripada sekadar “ini matcha”. Tawarkan juga pilihan ukuran serta panas atau dingin sebelum mengonfirmasi pesanan.
AKHIRI SESI PERCAKAPAN INI ?
Browser demo — your speech is transcribed by your browser, but the customer's replies and the grading here are canned. The real app runs both through an LLM.
Four features that turn a stack of documents into a confident shift.
Practice real customer conversations, out loud.
Live speech-to-text feeds a stateful, in-character AI customer whose reply is spoken back to you. Time-boxed 5-minute sessions in Bahasa Indonesia.
Upload real docs — get structured training.
The app extracts text from your café's PDF/DOCX on-device and an LLM rewrites it into clean, leveled, beginner-friendly lessons.
Coaching after every session, saved to share.
Each session ends with an AI summary and actionable suggestions. Every practice is saved and shareable with your manager via the iOS share sheet.
Sensitive SOPs stay protected and on-device.
Manager-only, permission-based access. Local auth with CryptoKit AES-GCM, and user data stays on the device — nothing stored on a server.
The roleplay escalates. Each persona is a real café archetype — beat one to face the next.
The Friendly Newcomer
A relaxed first-timer, curious about the menu and happy to ask for recommendations.
Tests: Product knowledge
The Rushing Auntie
Impatient and easily annoyed — interrupts and complains, but softens with fast, polite service.
Tests: Efficiency under pressure
The Perfectionist
Meticulous, with special requests and allergies; only satisfied by empathy and a real solution.
Tests: Active listening & problem-solving
The Entitled One
Demands fake discounts, compares to other cafés, threatens 1-star reviews — calmed only by polite, firm boundaries.
Tests: De-escalation & composure
A native iOS app, protocol-oriented and fully mockable.
PDFKit reads the text layer; scanned pages fall back to Vision OCR. DOCX is unzipped and its XML walked to preserve paragraphs.
An engineered Head Barista system prompt turns messy SOP/menu docs into warm, Markdown-structured, leveled lessons under a strict JSON contract.
A streaming loop: Apple Speech STT to a stateful in-character LLM to Gemini TTS, time-boxed to 5 minutes with a countdown and auto-feedback.
Register, login, and reset with passwords encrypted at rest via CryptoKit AES-GCM — fully on-device.
A WidgetKit + App Intents widget reads shared App Group state to ask trained today, backed by scheduled local notifications.
Lessons are generated once via API and stored in SwiftData, so the app works against previously generated content offline.
Made at the Apple Developer Academy.