ALTO
MVP Sprint

49 execution cards across 7 days. Monday March 2 – Sunday March 8, 2026. Click any card to cycle its state. Everything persists locally.

Overview

Sprint at a glance

8–10 hours daily. 42h estimated work + 14–28h buffer. Every card has a pivot plan.

Total Cards

~40h

Estimated

Days

Completed

Sprint Journal

Daily planning & reflection

Plan each day the night before. Reflect after. Write freely.

Monday March 2

Voice Pipeline Foundation

Talk into phone → see transcription → hear ElevenLabs speak it back

Goal	Voice echo loop working end-to-end
Planned hours
Actual hours
Cards done	0 / 7

Morning plan

Midday check-in

Evening reflection

Tuesday March 3

LLM Agent + Backend

Full voice conversation — talk to Alto, it reasons with GPT-4o-mini, it talks back

Goal	Voice → Backend → GPT-4o-mini → TTS loop
Planned hours
Actual hours
Cards done	0 / 8

Morning plan

Midday check-in

Evening reflection

Wednesday March 4

WhatsApp Integration

"Alto, what messages do I have?" → reads real WhatsApp → dictate reply → sent

Goal	Real WhatsApp messages by voice
Planned hours
Actual hours
Cards done	0 / 6

Morning plan

Midday check-in

Evening reflection

Thursday March 5

Gmail Integration

"Alto, read my latest emails" → reads real inbox → dictate reply → sent

Goal	Real Gmail by voice
Planned hours
Actual hours
Cards done	0 / 8

Morning plan

Midday check-in

Evening reflection

Friday March 6

Calendar + Briefing Engine

Start drive → Alto auto-briefs you on everything → handle it all by voice

Goal	All integrations live + briefing engine
Planned hours
Actual hours
Cards done	0 / 7

Morning plan

Midday check-in

Evening reflection

Saturday March 7

Driving Detection + UI + Onboarding

Get in car → Alto auto-activates → briefs → full voice control → onboarding works

Goal	Product-quality UI + auto-detection
Planned hours
Actual hours
Cards done	0 / 7

Morning plan

Midday check-in

Evening reflection

Sunday March 8

Ship + Demo

TestFlight link live + 30-second TikTok demo filmed

Goal	Ship it. Film it. Share it.
Planned hours
Actual hours
Cards done	0 / 6

Morning plan

Midday check-in

Evening reflection

Day 1 — Monday March 2

Voice Pipeline Foundation

Talk into phone → see transcription → hear ElevenLabs speak it back. The foundation of everything.

5.5h

Estimated

2.5–4.5h

Buffer

Cards

Critical risk: whisper.cpp integration — has a pivot to SFSpeechRecognizer if it fails.

1.1 Pending Est: 30min | Max: 45min

Xcode Project Setup

Who: You (Xcode manual)
Needs: Nothing — this is the starting block
Done when: App builds and runs on device — black screen shows "ALTO" circle with "Initializing..." text
Pivot: None — this must work

Notes

New Xcode project, iOS App, SwiftUI, iOS 17.0 min. Bundle ID: com.alto.app. Add Background Modes: Audio. Add Info.plist keys: Microphone, Motion, Location. Build to PHYSICAL device.

Comment

Start here. Get the blank canvas running. Don't overthink the project structure yet.

1.2 Pending Est: 20min | Max: 30min

Audio Session Manager

Who: Claude writes → You build + verify
Needs: 1.1
Done when: App shows "Audio ready" on screen (not "error")
Pivot: None — AVAudioSession is reliable

Notes

Category: .playAndRecord with .voiceChat mode. Options: defaultToSpeaker, allowBluetooth, allowA2DP. Must test on device.

Comment

Quick win. This just configures the audio hardware. Good momentum builder.

1.3 Pending Est: 45min | Max: 1.5h

Whisper Model + SPM Package

Who: Both — Claude downloads model + adds SPM, You verify
Needs: 1.1
Done when: SwiftWhisper package resolves without errors AND ggml-base.bin appears in Copy Bundle Resources
Pivot: If SwiftWhisper SPM fails after 45min, switch to SFSpeechRecognizer

Notes

SPM URL: github.com/exPHAT/SwiftWhisper, branch: master. Model from HuggingFace. Highest-risk step on Day 1.

Comment

This is the one that could bite you. If Xcode spins on "Resolving packages" for more than 15 min, force-quit and retry with clean derived data.

1.4 Pending Est: 1.5h | Max: 2h

Whisper Service (On-Device STT)

Who: Claude writes → You build + test on device
Needs: 1.2, 1.3
Done when: Tap circle → speak "Hello Alto" → tap again → your words appear as text on screen
Pivot: If transcription is garbage after 2h, switch to SFSpeechRecognizer fallback

Notes

Actor-isolated for thread safety. Records to [Float] buffer at 16kHz mono. Audio format must be pcmFormatFloat32, 16000Hz, 1 channel.

Comment

The "magic moment" of Day 1. When you see your words appear on screen from on-device Whisper, that's real.

1.5 Pending Est: 1h | Max: 1.5h

API Client + ElevenLabs TTS Service

Who: Claude writes → You test on device
Needs: 1.2
Done when: Hardcode a test string → hear ElevenLabs voice speak it through your phone speaker
Pivot: If ElevenLabs down → use AVSpeechSynthesizer (Apple TTS) as temp fallback

Notes

Model: eleven_flash_v2_5. Voice ID: 21m00Tcm4TlvDq8ikWAM (Rachel). Returns audio/mpeg → play via AVAudioPlayer. API key in code for now.

Comment

Have your ElevenLabs API key ready. The first time you hear a premium AI voice say something through your phone — that's the "wow" moment.

1.6 Pending Est: 30min | Max: 45min

Voice Echo Loop Test

Who: Both — Claude wires it up, You test on device
Needs: 1.4, 1.5
Done when: Full loop: tap → speak "Hello Alto" → tap → hear ElevenLabs say "You said: Hello Alto". Latency <3s.
Pivot: If audio conflicts → check AudioSession category

Notes

Echo mode — no LLM yet, just speech → text → speech. Tests the full audio pipeline end-to-end.

Comment

Your Day 1 demo moment. Record a quick video of this working — it's your first "building in public" content.

1.7 Pending Est: 10min | Max: 15min

Git Init + First Commits

Who: Claude (terminal)
Needs: 1.6
Done when: git log shows 3 commits: project setup, Whisper STT, ElevenLabs TTS + echo loop
Pivot: None

Notes

git init in /Desktop/Alto/. Don't commit ggml-base.bin (150MB) — add to .gitignore.

Comment

End of day. Clean commits. You're done. Go celebrate — you have a working voice pipeline.

Day 2 — Tuesday March 3

LLM Agent + Backend

Full voice conversation — talk to Alto, it reasons with GPT-4o-mini, it talks back.

Estimated

2–4h

Buffer

Cards

Critical risk: OpenAI API key + Cloudflare deploy — both should be smooth.

2.1PendingEst: 30min | Max: 45min

Cloudflare Workers Project Setup

Who: Claude (terminal)
Needs: Nothing (parallel track)
Done when: alto-api/ directory with wrangler.toml, package.json, correct folder structure
Pivot: None — standard setup

Notes

Create inside /Desktop/Alto/alto-api/. Pages project for file-based routing.

Comment

Claude handles this entirely. You just watch.

2.2PendingEst: 20min | Max: 30min

D1 Database + Schema

Who: Claude (terminal)
Needs: 2.1
Done when: wrangler d1 execute returns: users, actions_log, drive_sessions
Pivot: None

Notes

3 tables: users, actions_log, drive_sessions. Users stores OAuth tokens + Unipile account ID.

Comment

Quick. Database ready in minutes.

2.3PendingEst: 30min | Max: 45min

JWT Auth Helpers

Who: Claude writes
Needs: 2.1
Done when: _auth.js exports signJWT() and verifyJWT() that correctly sign and verify HMAC-SHA256 tokens
Pivot: None — pure crypto, well-tested pattern

Notes

Web Crypto API (native to Workers). HMAC-SHA256 with JWT_SECRET from env. Token expiry: 24h.

Comment

Auth + middleware in one shot. Standard stuff.

2.4PendingEst: 30min | Max: 45min

Auth Endpoints (Register + Login)

Who: Claude writes
Needs: 2.2, 2.3
Done when: curl register → JWT token. curl login → JWT token. Wrong password → 401.
Pivot: None

Notes

Password hashing: PBKDF2 with random salt (Web Crypto). Register returns token immediately (auto-login).

Comment

After this card, you have a real backend with user accounts.

2.5PendingEst: 1.5h | Max: 2h

Agent Chat Endpoint (GPT-4o-mini + Tools)

Who: Claude writes
Needs: 2.3, 2.4
Done when: curl with "What messages do I have?" returns intelligent response + mock tool calls
Pivot: If OpenAI errors → check API key. If tool-calling broken → add few-shot examples.

Notes

THIS IS THE BRAIN OF ALTO. 6 tool definitions. Day 2 uses MOCK tool responses. Supports tool-call loops. Logs all actions.

Comment

Most important card of the sprint. Take your time getting the system prompt right.

2.6PendingEst: 30min | Max: 45min

Set Secrets + Deploy + Test Backend

Who: Both — Claude deploys, You verify with curl
Needs: 2.5
Done when: All 3 curl commands succeed: register, login, chat with token
Pivot: If deploy fails → check wrangler.toml config

Notes

Secrets: JWT_SECRET, OPENAI_API_KEY, ELEVENLABS_API_KEY. Deploy: wrangler pages deploy public.

Comment

Moment of truth for the backend. All three curls working = green light to wire up iOS.

2.7PendingEst: 1h | Max: 1.5h

iOS Agent + API Client Chat Method

Who: Claude writes → You build
Needs: Day 1 complete, 2.6
Done when: AltoAgent.swift + APIClient chat method compile without errors
Pivot: None — standard Swift networking

Notes

AltoAgent manages conversation history (last 20 msgs). APIClient.chat() sends text + history → gets response.

Comment

Bridging iOS and backend. Straightforward networking code.

2.8PendingEst: 1h | Max: 1.5h

Full Voice Conversation Loop Test

Who: Both — Claude wires ContentView, You test on device
Needs: 2.7
Done when: On device: tap → say "What's on my calendar?" → tap → hear Alto respond via ElevenLabs. Latency <4s.
Pivot: If latency >6s → add timing logs to find bottleneck

Notes

Circle colors: red (listening), purple (thinking), blue (speaking), white (ready). Mock tools return fake data — expected.

Comment

THE DEMO MOMENT OF DAY 2. When you have a real conversation with Alto — voice in, AI thinks, voice out — that's the product.

Day 3 — Wednesday March 4

WhatsApp Integration

"Alto, read my messages" → hears real WhatsApp messages → dictates reply → sent.

Estimated

3–5h

Buffer

Cards

Critical risk: Unipile connection stability + WhatsApp pairing. If Unipile is down, skip to Day 4 (Gmail).

3.1PendingEst: 45min | Max: 1.5h

Unipile Setup + WhatsApp Pairing

Who: You (Unipile dashboard + phone)
Needs: 2.6 (backend deployed)
Done when: Unipile dashboard shows WhatsApp "connected". Test API call returns real messages.
Pivot: If pairing fails → skip to Day 4 (Gmail) and come back later

Notes

You need Unipile DSN + API key (from Drifo). Set secrets: UNIPILE_DSN and UNIPILE_API_KEY.

Comment

This is the gate. If Unipile connects, the rest of Day 3 is smooth.

3.2PendingEst: 45min | Max: 1h

WhatsApp Messages Endpoint

Who: Claude writes
Needs: 3.1
Done when: curl /api/whatsapp/messages returns real WhatsApp messages with sender name + text + timestamp
Pivot: If Unipile API format changed → check their docs. Reuse Drifo patterns.

Notes

GET /api/whatsapp/messages?contact=Mike&limit=10. Maps Unipile response to clean format.

Comment

Proven pattern from Drifo. Should be quick.

3.3PendingEst: 45min | Max: 1h

WhatsApp Send Endpoint

Who: Claude writes
Needs: 3.1
Done when: curl /api/whatsapp/send → message actually appears in the contact's WhatsApp
Pivot: If send fails → check chat_id resolution

Notes

POST /api/whatsapp/send { contact, message }. Searches recent chats by contact name to find chat_id.

Comment

Test with a safe contact first. Send yourself a message via the API.

3.4PendingEst: 30min | Max: 45min

Wire Real WhatsApp Tools Into Agent

Who: Claude writes
Needs: 3.2, 3.3
Done when: curl agent/chat with "Read my WhatsApp messages" returns REAL messages, not mock data
Pivot: If internal calls fail → check getInternalToken() and getBaseURL()

Notes

Replace mock cases in executeTool() with real fetch calls. Internal calls use short-lived JWT (60s expiry).

Comment

The moment where mock → real. When the agent reads your ACTUAL WhatsApp messages — that's real.

3.5PendingEst: 30min | Max: 45min

Deploy + Test WhatsApp E2E

Who: Claude deploys → You verify with curl
Needs: 3.4
Done when: Backend deployed. curl chat endpoint with WhatsApp queries returns real data.
Pivot: None at this point

Notes

Test: "What messages do I have?" and "Send Mike a message saying I'll be 10 minutes late".

Comment

Backend checkpoint. Everything works via curl.

3.6PendingEst: 1h | Max: 1.5h

Voice Test: WhatsApp by Voice

Who: You (on device)
Needs: 3.5
Done when: On phone: "Alto, what WhatsApp messages do I have?" → hear real messages → "Reply to Mike saying I'm on my way" → message sent
Pivot: If voice works but messages wrong → check backend via curl independently

Notes

No iOS code changes needed — purely a testing/verification card. Test the confirmation flow.

Comment

Day 3 demo moment. Real WhatsApp, real voice. Film this. This is TikTok content.

Day 4 — Thursday March 5

Gmail Integration

"Alto, read my latest emails" → reads real inbox → dictate reply → sent.

6.5h

Estimated

1.5–3.5h

Buffer

Cards

Critical risk: Google OAuth flow — needs careful setup in Google Cloud Console.

4.1PendingEst: 1h | Max: 1.5h

Google Cloud Console Setup

Who: You (browser — Google Cloud Console)
Needs: Nothing
Done when: OAuth consent screen configured + OAuth 2.0 Client ID created with correct redirect URI
Pivot: Use "Testing" mode (100 users max) — fine for MVP

Notes

Enable Gmail API + Calendar API. Scopes: gmail.modify, calendar.events. Bundle ID: com.alto.app. Redirect URI: com.alto.app:/oauth2redirect.

Comment

Most tedious card of the sprint. Google's console is a maze. Take your time.

4.2PendingEst: 30min | Max: 45min

Google OAuth Backend Endpoint

Who: Claude writes
Needs: 4.1
Done when: /api/auth/google endpoint accepts auth code, exchanges for tokens, stores in D1
Pivot: If token exchange fails → double-check redirect_uri match

Notes

POST /api/auth/google { code, redirect_uri }. Set secrets: GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET.

Comment

Standard OAuth token exchange. Quick card.

4.3PendingEst: 20min | Max: 30min

Token Refresh Helper

Who: Claude writes
Needs: 4.2
Done when: refreshGoogleToken() uses refresh_token to get new access_token when expired
Pivot: None

Notes

Added to _auth.js. Called by Gmail + Calendar endpoints before every request. Google tokens expire after 1h.

Comment

Small but critical. Without this, Gmail breaks after 1 hour.

4.4PendingEst: 1h | Max: 1.5h

iOS Google OAuth Flow

Who: Claude writes → You test on device
Needs: 4.2
Done when: On device: tap "Connect Google" → sign in → redirects back → backend stores tokens
Pivot: If ASWebAuthenticationSession doesn't redirect → check URL scheme in Info.plist

Notes

Uses ASWebAuthenticationSession. URL scheme: com.alto.app. Scopes: gmail.modify + calendar.events.

Comment

OAuth on mobile is always fiddly. Triple-check redirect URIs match character for character.

4.5PendingEst: 45min | Max: 1h

Gmail Messages Endpoint

Who: Claude writes
Needs: 4.3
Done when: curl /api/gmail/messages returns real unread emails with sender, subject, snippet, date
Pivot: If Gmail API returns 403 → check scopes in consent screen

Notes

Two-step: list message IDs, then fetch metadata. Headers: From, Subject, Date + snippet. Uses refreshGoogleToken().

Comment

Seeing your real emails come through the API is satisfying.

4.6PendingEst: 30min | Max: 45min

Gmail Send Endpoint

Who: Claude writes
Needs: 4.3
Done when: curl /api/gmail/send → email actually arrives in recipient's inbox
Pivot: If send fails → check RFC 2822 formatting / base64url encoding

Notes

POST /api/gmail/send { to, subject, body, thread_id? }. Composes RFC 2822 email, base64url encodes.

Comment

Send to your own email first. Verify it lands.

4.7PendingEst: 20min | Max: 30min

Wire Gmail Tools Into Agent

Who: Claude writes
Needs: 4.5, 4.6
Done when: curl agent/chat with "Read my emails" returns real email data (not mock)
Pivot: None — same pattern as WhatsApp wiring

Notes

Replace mock read_email and send_email in executeTool(). Same pattern as WhatsApp.

Comment

Copy-paste pattern from Day 3.4. Fast.

4.8PendingEst: 1h | Max: 1.5h

Deploy + Voice Test: Email by Voice

Who: Both — Claude deploys, You test on device
Needs: 4.7, 4.4
Done when: On phone: "Alto, read my latest emails" → hear real inbox → "Reply to Lisa saying I'll review tonight" → email sent
Pivot: If email reads but won't send → test send endpoint with curl

Notes

Deploy backend first, then test voice. Test flow: read → confirm → send → verify.

Comment

Day 4 done. WhatsApp + Gmail both working by voice. You're holding a serious product.

Day 5 — Friday March 6

Calendar + Briefing Engine

Start drive → Alto auto-briefs you → handle everything by voice.

5.5h

Estimated

2.5–4.5h

Buffer

Cards

Low risk — Calendar API is simpler than Gmail. The briefing engine is the creative piece.

5.1PendingEst: 30min | Max: 45min

Calendar Events Endpoint

Who: Claude writes
Needs: 4.3 (token refresh)
Done when: curl /api/calendar/events returns today's real calendar events
Pivot: If no events → check timezone (API is timezone-sensitive)

Notes

Returns: title, start, end, location, description snippet. Timezone: Europe/Berlin (hardcoded for MVP).

Comment

Easiest integration day. Calendar API is clean.

5.2PendingEst: 30min | Max: 45min

Calendar Create Endpoint

Who: Claude writes
Needs: 4.3
Done when: curl /api/calendar/create → event appears in your Google Calendar
Pivot: None

Notes

POST /api/calendar/create { title, date, time, duration }. Default duration: 60 min. Primary calendar.

Comment

Create a test event and check your calendar app.

5.3PendingEst: 20min | Max: 30min

Wire Calendar Tools Into Agent

Who: Claude writes
Needs: 5.1, 5.2
Done when: ALL mock tools replaced. ZERO mock data. curl agent/chat with "What's on my calendar?" returns real events.
Pivot: None

Notes

After this card, all 6 tools are live. Everything real.

Comment

Milestone. Alto can now actually DO things in the real world via voice.

5.4PendingEst: 1h | Max: 1.5h

Briefing Engine Endpoint

Who: Claude writes
Needs: 5.3 (all tools real)
Done when: curl /api/agent/briefing returns a natural, conversational 3-5 sentence summary of WhatsApp, email, and calendar
Pivot: If briefing sounds robotic → tune the prompt. If too long → add max_tokens=200.

Notes

Fetches WhatsApp (10), Gmail unread (5), Calendar in parallel. Priority: calendar first, then messages needing reply, then FYI.

Comment

This is what makes Alto special. Spend time on the prompt.

5.5PendingEst: 30min | Max: 45min

iOS Briefing Service

Who: Claude writes → You build
Needs: 5.4
Done when: BriefingService.swift compiles. fetchBriefing() returns the briefing text.
Pivot: None — simple HTTP call

Notes

@Observable class with hasBriefed state. Called once when drive starts. reset() for new drives.

Comment

Simple networking wrapper. Quick card.

5.6PendingEst: 1h | Max: 1.5h

Deploy + E2E Test: All Integrations

Who: Both — Claude deploys, You test on device
Needs: 5.5
Done when: All 6 tools work by voice: WhatsApp read/send, Gmail read/send, Calendar read/create
Pivot: If any integration broke → test that endpoint with curl first

Notes

Comprehensive integration test. Test all 6 tools by voice. Also test the briefing on app restart.

Comment

If all 6 pass, you've built the core product. Record a mega-demo video. Celebrate.

5.7PendingEst: 10min | Max: 15min

Commit: All Integrations Complete

Who: Claude (terminal)
Needs: 5.6
Done when: Clean commits for calendar + briefing. git log tells the story.
Pivot: None

Notes

"feat: Google Calendar integration" and "feat: drive-start briefing engine".

Comment

End of the "hard" part. Days 6-7 are polish. The core product is DONE.

Day 6 — Saturday March 7

Driving Detection + UI + Onboarding

Get in car → Alto auto-activates → briefs → full voice control → onboarding works.

Estimated

2–4h

Buffer

Cards

Low risk — all features are local iOS, no API dependencies.

6.1PendingEst: 1h | Max: 1.5h

Driving Detector (CoreMotion + GPS)

Who: Claude writes → You test in car
Needs: Day 1 iOS foundation
Done when: On device: start driving → state changes to .driving. Park for 2min → changes to .parked.
Pivot: If CoreMotion unreliable → rely purely on GPS speed (>15 km/h = driving)

Notes

State machine: .idle → .driving → .parked. Uses CLLocationManager + CMMotionActivityManager.

Comment

You'll need to actually drive to test this. A short trip around the block is enough.

6.2PendingEst: 20min | Max: 30min

Driving Session Manager

Who: Claude writes
Needs: 6.1
Done when: DrivingSession tracks isActive, startTime, actionsCount, durationFormatted
Pivot: None — simple state class

Notes

@Observable class. start(), end(), recordAction(). durationFormatted: "23 min".

Comment

Quick bookkeeping class. 5 minutes of code.

6.3PendingEst: 1.5h | Max: 2h

Main View (Pulsing Circle UI)

Who: Claude writes → You review + iterate
Needs: 6.1, 6.2, Day 5 complete
Done when: Beautiful minimal UI: pulsing circle changes color, status text, activity log, "End Drive" button, auto-briefing plays
Pivot: If animations jank → simplify to opacity changes only

Notes

Pure black background. Circle: white (ready), red (listening), purple (thinking), blue (speaking). Auto-briefing on .task { }.

Comment

This is where it stops looking like a dev test and starts looking like a PRODUCT.

6.4PendingEst: 1h | Max: 1.5h

Onboarding View

Who: Claude writes → You test on device
Needs: 2.4 (auth endpoints)
Done when: Fresh install → onboarding → enter email + password → register → lands on MainView
Pivot: If register fails → test backend auth with curl independently

Notes

Minimal: ALTO title + tagline + email + password. Register-first flow.

Comment

Keep it dead simple. Email + password. Get in. The product is the voice, not the signup form.

6.5PendingEst: 20min | Max: 30min

Content View Router

Who: Claude writes
Needs: 6.3, 6.4
Done when: App flow: not logged in → OnboardingView → login → MainView. Token persists across restarts.
Pivot: None

Notes

ContentView checks for saved token. If token: MainView. If not: OnboardingView.

Comment

Simple navigation. Quick card.

6.6PendingEst: 30min | Max: 45min

Keychain Token Storage

Who: Claude writes → You test
Needs: 6.4, 6.5
Done when: Token survives app kill + restart. API keys no longer hardcoded.
Pivot: If Keychain flaky → use UserDefaults for now

Notes

Store: JWT token, ElevenLabs API key, Google client ID. Remove all hardcoded keys from source.

Comment

Security cleanup. No more "YOUR_API_KEY" in code.

6.7PendingEst: 1h | Max: 1.5h

Full Integration Test

Who: You (on device, ideally in car)
Needs: 6.6
Done when: Full user journey: fresh install → onboarding → briefing → voice control → all integrations → "End Drive"
Pivot: List all bugs, prioritize for Day 7

Notes

Delete app, reinstall, go through onboarding. Test in car if possible. Note every bug/rough edge.

Comment

Pretend you're a new user. What's confusing? What breaks? What's amazing?

Day 7 — Sunday March 8

Ship + Demo

TestFlight link live + 30-second demo filmed. Focus on "good enough to show", not perfection.

5.5h

Estimated

2.5–4.5h

Buffer

Cards

This is ship day. The demo IS the product-market fit test.

7.1PendingEst: 45min | Max: 1h

Post-Drive Summary View

Who: Claude writes → You review
Needs: 6.2 (DrivingSession)
Done when: After "End Drive" → see summary with action count, drive duration, activity log
Pivot: If time is tight → skip entirely. Nice-to-have.

Notes

Shows: # actions, drive time, scrollable log. Minimal design: monochrome, monospaced.

Comment

Satisfying to see after a drive. But if you're behind schedule, skip it.

7.2PendingEst: 2h | Max: 2.5h

Error Handling + Edge Cases

Who: Claude writes → You test
Needs: Day 6 complete
Done when: App doesn't crash on: no internet, empty transcription, backend error, token expired, model load fail
Pivot: Fix crashes first. Polish error messages last.

Notes

Focus on network failures, empty states, auth expiry. Add loading states and retry logic (1 retry, no loop).

Comment

Bug bash. Go through every path that could fail.

7.3PendingEst: 45min | Max: 1h

Voice Interruption Handling

Who: Claude writes → You test
Needs: 6.3
Done when: While Alto is speaking, tap → TTS stops → mic starts → you can speak immediately
Pivot: If audio session conflicts → add a 200ms delay between stopping TTS and starting mic

Notes

ttsService.stop() first, then startListening(). AudioSession should handle the switch automatically.

Comment

UX polish that makes it feel real. Being able to interrupt Alto mid-sentence is natural.

7.4PendingEst: 30min | Max: 45min

App Icon

Who: You (design tool or Claude generates)
Needs: Nothing
Done when: App icon shows on home screen + TestFlight. Clean, monochrome, recognizable.
Pivot: Use a simple "A" lettermark in white on black

Notes

Simple monochrome. Need 1024x1024 for App Store. Add to Assets.xcassets → AppIcon.

Comment

Keep it simple. A clean icon on black. Done.

7.5PendingEst: 1h | Max: 1.5h

TestFlight Build + Upload

Who: You (Xcode)
Needs: 7.2, 7.4
Done when: TestFlight email arrives. Install Alto from TestFlight on a clean device and full flow works.
Pivot: If archive fails → check signing + capabilities

Notes

Product → Archive → Distribute → App Store Connect. Version 1.0 (1). Wait ~15-30 min after upload.

Comment

Standard TestFlight flow. First build takes longest because of signing setup.

7.6PendingEst: 1.5h | Max: 2h

Film Demo Video

Who: You (in car, filming)
Needs: 7.5
Done when: 30-second TikTok-ready demo: phone on mount, auto-briefing, voice commands, hands on wheel
Pivot: If in-car filming doesn't work → film parked in driveway. Voice-over with screen recording as backup.

Notes

Dashcam POV or passenger-seat angle. Phone visible but never touched. 30 seconds MAX. Caption: "I built an AI co-driver in 7 days".

Comment

This is your launch moment. The video sells Alto better than any landing page. Don't skip this card.

Sprint Summary

Fill Sunday evening

How did it go?

Total hours worked
Cards completed	0 / 49
Days on schedule
Biggest win
Biggest surprise
What I'd do differently
Next priorities

Appendix A

Pre-Sprint Checklist

Complete these by Sunday March 1 evening so Monday starts clean.

ElevenLabs API key — sign up at elevenlabs.io, get API key
OpenAI API key — verify your key works, has GPT-4o-mini access
Unipile credentials — confirm UNIPILE_DSN + UNIPILE_API_KEY from Drifo
Google Cloud Console — could start Task 4.1 early (OAuth setup takes 1h)
Physical iOS device — charged, connected to Xcode, trusted
USB cable — for tethered debugging
Xcode updated — latest version, iOS 17+ SDK
wrangler CLI — npm install -g wrangler, wrangler login
Camera/mount for demo day — phone mount for car filming

Appendix B

Secrets Required

Secret	Where	When
ELEVENLABS_API_KEY	iOS code + Cloudflare	Day 1
OPENAI_API_KEY	Cloudflare	Day 2
JWT_SECRET	Cloudflare	Day 2
UNIPILE_DSN	Cloudflare	Day 3
UNIPILE_API_KEY	Cloudflare	Day 3
GOOGLE_CLIENT_ID	Cloudflare + iOS	Day 4
GOOGLE_CLIENT_SECRET	Cloudflare	Day 4

Appendix C

Card Count Summary

Day	Focus	Cards	Est. Hours	Buffer
Day 1	Voice Pipeline	7	5.5h	2.5–4.5h
Day 2	LLM + Backend	8	6h	2–4h
Day 3	WhatsApp	6	5h	3–5h
Day 4	Gmail	8	6.5h	1.5–3.5h
Day 5	Calendar + Briefing	7	5.5h	2.5–4.5h
Day 6	UI + Onboarding	7	6h	2–4h
Day 7	Ship + Demo	6	5.5h	2.5–4.5h
Total		49	~40h	16–30h

Appendix D

Fallback Decision Tree

When things go wrong, follow the pivot. Don't burn the day.

whisper.cpp won't compile (>1.5h)
Switch to SFSpeechRecognizer (Apple's built-in STT). Worse accuracy but zero integration friction. Can always swap back to Whisper later.

ElevenLabs API down or too slow
Use AVSpeechSynthesizer (Apple TTS) for dev/testing. Switch back when it's up. Or try OpenAI TTS API as alternative.

Unipile WhatsApp won't connect
Skip Day 3 entirely → do Day 4 (Gmail) instead. Come back on Day 6 buffer time. Worst case: ship without WhatsApp, add in v1.1.

Google OAuth consent screen blocked
Use Testing mode (100 users max) — fine for MVP. Add your email as test user. Submit for verification in parallel.

GPT-4o-mini tool-calling unreliable
Add few-shot examples to system prompt. If still bad → try gpt-4o (more expensive but better). Last resort: Claude Haiku (proven in Drifo).

Can't finish Day 6-7 features
Core loop works by Day 5. Day 6-7 are polish — skip driving detection + summary. Ship with manual tap-to-activate instead of auto-detect.

ALTOMVP Sprint

Sprint at a glance

Daily planning & reflection

Voice Pipeline Foundation

LLM Agent + Backend

WhatsApp Integration

Gmail Integration

Calendar + Briefing Engine

Driving Detection + UI + Onboarding

Ship + Demo

Voice Pipeline Foundation

LLM Agent + Backend

WhatsApp Integration

Gmail Integration

Calendar + Briefing Engine

Driving Detection + UI + Onboarding

Ship + Demo

Fill Sunday evening

Pre-Sprint Checklist

Secrets Required

Card Count Summary

Fallback Decision Tree

ALTO
MVP Sprint