WristCall AI

Just speak. WristCall AI makes the call, talks on your behalf, and brings the answer straight to your wrist.

Comment

⌚ WristCall AI — Devpost Hackathon Story & Technical Showcase

**Empowering Smartwatches to Search, Discover, Call, and Converse

GitHub Repository Wear OS Jetpack Compose Kotlin SerpApi Groq AI CALL-E API TextToSpeech

🚨 The Problem We Are Solving

Despite massive advances in smartwatches, making a simple phone call to a real-world business is still surprisingly high-friction.

1. ⌚ The Wearable Inaction Gap

Smartwatches are excellent at displaying information and notifications, but they are still limited when users need to take real-world action—such as checking business hours, asking about prices, confirming availability, or making an inquiry.

2. 📱 Phone Number Friction

Users rarely memorize local business phone numbers. Searching Google and typing on a tiny smartwatch screen is slow, frustrating, and error-prone.

3. ⏳ Time Wasted on Routine Calls

Calling a business often means navigating IVR menus, waiting on hold, explaining the same request, and spending several minutes just to get a simple answer.

4. 🧠 No Useful Record of the Conversation

Even after completing a call, users are often left to remember what was discussed. There is no convenient, structured summary waiting on their wrist.

💡 The Solution: WristCall AI

WristCall AI turns a Wear OS smartwatch into an autonomous voice call delegate.

Simply speak one sentence on your wrist, and WristCall AI handles the rest:

🎙️ Speak → 🔍 Find → 🧠 Understand → 📞 Call → ✨ Summarize

WristCall AI can discover publicly available business phone numbers through SerpApi, understand and refine the user's request using Groq AI, and delegate the actual conversation through CALL-E.

Instead of manually making routine calls, users can simply say:

"Call the barber and ask for the earliest appointment tomorrow."

or

"Call the doctor's clinic and ask if they have appointments available this week."

or

"Call the restaurant and ask if they have a table available tonight."

WristCall AI handles the conversation and returns the important information directly to the user's wrist through a concise 3-bullet executive summary and voice readout.

🌎 From Your Wrist to the Real World

Whether it's a barber, doctor, restaurant, hotel, repair shop, garage, store, or any other business with a publicly available phone number, WristCall AI aims to turn a simple voice command into a completed real-world phone task—without requiring the user to pull out their phone, search for a number, or personally handle the entire conversation.

💡 Inspiration

In today's fast-paced world, smartwatches excel at notifications and health tracking, but when it comes to taking real action in the real world—like calling a local restaurant to check opening hours, inquiring if a auto shop has a repair slot, or finding out if a BBQ place still has best sellers left—users are forced to pull out their phones, search Google, find phone numbers, dial manually, and wait on hold.

We asked ourselves:
"What if your smartwatch could autonomously search Google, synthesize intent, dial real phone numbers, hold human-like conversations, and present you with a 3-bullet executive summary—all from your wrist?"

That vision inspired WristCall AI—bringing autonomous AI phone call delegation straight to Wear OS.

⚡ What It Does

WristCall AI turns your smartwatch into an autonomous voice call delegation assistant:

  1. 🎙️ Zero-Friction Voice & Hardware Stem Activation: Double-press your smartwatch physical side crown button or tap the central mic button to activate voice listening with a live 3-word sliding marquee ticker ([Word 1] [Word 2] [Word 3]).
  2. 🔍 Automatic Business Phone & Snippet Discovery (Google SerpApi): Users don't need to look up or memorize phone numbers. Simply say "Call Cattleack Barbeque in Farmers Branch and ask opening hours". WristCall AI queries Google via SerpApi to extract verified E.164 phone numbers, store addresses, and location snippets.
  3. 🧠 Groq AI 120B Intent Synthesis (gpt-oss-120b): Combines raw user prompts with Google search snippets to synthesize structured, high-precision call instructions.
  4. 📞 CALL-E Telecom Dispatch & Live Polling: Dispatches real autonomous phone calls via CALL-E REST API (/v1/calls), tracking live status (Queued $\rightarrow$ Ringing $\rightarrow$ In Conversation $\rightarrow$ Completed) on the watch UI.
  5. ✨ Dedicated 3-Bullet Executive Summary & Audio Playback: Displays a glowing 3-bullet summary screen with automatic TextToSpeech voice readout and a 🔊 Listen Recording button to play real audio call recordings.
  6. 📱 2 Editable Quick-Action Cards: Persistent home screen cards with inline ✏️ Edit buttons for 1-click repeat task delegation saved to local storage.

🛠️ How We Built It

🏗 System Architecture Diagram

graph TD
    A[⌚ Wear OS Smartwatch] -->|Hardware Stem / Mic Touch| B[WristCall Wear App UI]
    B -->|User Instruction| C{Phone Number Provided?}
    C -->|No - Business Name Only| D[🔍 Google SerpApi Client]
    D -->|Verified Phone + Snippets| E[🧠 Groq AI 120B Synthesis]
    C -->|Yes - Direct E.164| E
    E -->|Refined Prompt + E.164| F[📡 CALL-E REST Client]
    F -->|Dispatch /v1/calls| G[📞 CALL-E Telecom Network]
    G -->|Real-world Phone Call| H[🏪 Target Business / Recipient]
    H -->|Live Transcript & Audio| G
    G -->|JSON Response & Poll| B
    B -->|Render Screen + Audio Readout| I[📜 Transcript & ✨ 3-Bullet Summary Screen]

🎨 Design System & Color Palette

We built a custom dark slate & neon mint aesthetic specifically tailored for circular Wear OS displays:

Element Color Hex Visual Purpose
Background Slate #0F172A Deep contrast for circular OLED smartwatch displays
Active Mint Accent #6EE7B7 / #059669 Highlight badges, success states, & positive actions
Pulsing Mic Active #EF4444 / #991B1B Live recording indication & hardware marquee focus
Indigo Accent #6366F1 / #312E81 Settings gear, API keys status, & card container
Text Primary #F8FAFC / #E2E8F0 High-legibility typography

🛠️ Tech Stack & SDK Integration

Component Technology / Service Description
Wear OS UI Jetpack Compose for Wear OS Modern declarative UI components tailored for circular watch screens
Core Architecture Kotlin Coroutines & Flows Asynchronous API polling, timer states, & non-blocking UI threads
Google Phone Discovery Google SerpApi Engine Discovers verified business phone numbers, addresses, & snippets
AI Intent Synthesis Groq AI Chat API (gpt-oss-120b) Synthesizes Google search context into structured voice instructions
Telecom Call Dispatch CALL-E REST API (/v1/calls) Executes autonomous phone calls to real-world phone networks
Voice & Audio Engine Android TextToSpeech & MediaPlayer Automatic summary voice readout & call recording audio playback
Local Persistence Android SharedPreferences Stores API keys, custom prompt cards, & user preferences

🚨 Challenges We Ran Into

  1. Wear OS Emulator RAM & Heap Limitations:

    • Challenge: Wear OS emulators default to 512 MB RAM and 48 MB Heap, which caused severe Garbage Collection (GC) thrashing and input dispatch timeouts (ANRs) when rendering Jetpack Compose layouts.
    • Solution: Upgraded the Wear_Jp emulator config.ini to 2048 MB RAM and 256 MB VM Heap, resulting in smooth, lag-free execution.

  2. Google Search Phone Number Extraction:

    • Challenge: Raw search snippets often contain formatted phone numbers ((972) 805-0999 or 972.805.0999) that fail strict E.164 formatting requirements.
    • Solution: Built a robust regex sanitizer in SmartCallResolver.kt that extracts, formats, and validates strings into clean +1XXXXXXXXXX E.164 phone numbers.

  3. CALL-E Status Polling & Dialogue Parsing:

    • Challenge: Decoding nested turn-by-turn dialogue attempts (recipients[0].attempts[0].transcript_turns) while handling live call status polling.
    • Solution: Developed CallEClient with safe JSON parsing and fallback summary generation to ensure transcripts render reliably even if telecom attempts fail or time out.

  4. Hardware Stem Double-Click Interception:

    • Challenge: Handling physical side crown key events (KEYCODE_STEM_PRIMARY) without hijacking system back navigation.
    • Solution: Implemented a 50ms–650ms key time-delta detector in MainActivity.kt, explicitly excluding KEYCODE_BACK from double-click triggers.

🏆 Accomplishments That We're Proud Of

📦 Open-Source calle-android-sdk: We open-sourced and published the official Kotlin/Android SDK on JitPack (com.github.Baklolman69:calle-android-sdk:1.0.0) and submitted PR #381 to CALLE-AI/awesome-phone-call-agents so any Android developer can integrate CALL-E into their app in under 5 minutes.

  • 🚀 Sub-Second Hardware Launch: Users can double-click their smartwatch side button to launch into mic listening mode in under 1 second.
  • 🔍 Zero-Phone-Number Phone Discovery: Users only need to name a business—the app handles searching Google, finding numbers, and building call prompts autonomously.
  • 🎙️ Right-to-Left Live Speech Marquee: Built a custom 3-word sliding marquee window that brings visual excitement to voice recording.
  • ✨ Executive 3-Bullet AI Summary & Voice Readout: Beautiful neon mint Wear OS executive summary screen with automated voice readout.
  • 🔑 Out-of-the-Box Zero-Friction Setup: Pre-configured with default CALL-E, SerpApi, and Groq AI credentials so judges can test instantly without manual setup.

🎓 What We Learned

  • Wear OS UX Design Principles: Less is more on circular displays. Designing for quick touches, large tap targets, and visual feedback makes smartwatch apps feel magical.
  • Combining SerpApi + Groq AI + CALL-E: Chaining Google search context with fast LLM inference (gpt-oss-120b) creates far more reliable phone call prompts than sending raw user input directly to telecom networks.
  • Resilient Mobile SDK Architecture: Separating the core logic into an independent calle-android SDK module ensured full reusability between phone and Wear OS modules.

🚀 What's Next for WristCall AI

We are excited to expand WristCall AI into a full-fledged autonomous wearable agent. Our upcoming roadmap includes:

1. 🎙️ "ELA" Always-On Wake-Word Voice Assistant

  • Hands-Free Wake-Word Detection: Users won't even need to touch or double-click the watch screen. Simply say "Hey ELA" followed by your command ("Hey ELA, call Pizza Hut and check delivery time").
  • On-Device Keyword Spotting: Lightweight TensorFlow Lite wake-word engine running locally on Wear OS for zero-latency background activation.

2. 💳 Autonomous In-App Wallet & Payment System

  • Hands-Free Deposits & Payments: Integrated in-app wallet where users deposit funds (via Google Pay / Stripe).
  • Autonomous Booking & Purchase Execution: When WristCall AI calls a business for pre-bookings, table reservations, or order deposits, CALL-E can securely authorize payments on the user's behalf up to a user-defined limit (e.g. paying a $20 deposit for a BBQ reservation or paying for a pizza pickup order).

3. 📅 Calendar & Contacts Deep Integration

  • Automatic calendar event creation when a phone call confirms an appointment or reservation time.
  • Direct sync with Android Contacts for calling personal and business contacts hands-free.

Built With

  • google-serach
  • groq
  • kotlin
  • serpapi
  • wearos
Share this project:

Updates

Submission history