Build message generation and scheduling system

[BillChirico]

💬 Build Message Generation and Scheduling System

Overview

Create the engine that decides when, what, and how to message users — personalized, timely, meaningful.

---

Goals

• Right message at the right time
• Context-aware (recent check-ins, triggers, milestones)
• Personalized based on learned patterns
• Never generic or robotic

---

Message Types

Type	Trigger	Example
Check-in	Daily schedule	"Hey, how's today going?"
Anticipatory	Pattern prediction	"Friday night coming up. Want to lock in your plan?"
Contextual	Recent event	"Rough day yesterday — you good?"
Milestone	Streak achieved	"30 days. That's not luck — that's you."
Supportive	Slip reported	"Slips happen. You're not starting over."
Tool suggestion	High-risk moment	"Want to try a breathing exercise?"

---

Scheduling Logic

Schedule Message:
  IF quiet hours → skip
  IF user disabled Buddy → skip
  IF recent message (< 4 hours) → skip
  
  Determine optimal time:
    - Learned check-in window
    - User timezone
    - No quiet hours
    
  Generate message:
    - Context from mem0
    - User tone preference
    - Current situation (triggers, milestones)
    
  Send via push notification

---

Message Generation Pipeline

1. Context Assembly — Gather from mem0 (patterns, history)
2. Prompt Building — Create AI prompt with context
3. Generation — Vercel AI SDK streaming response
4. Personalization — Apply tone, insert user details
5. Delivery — Push notification with deep link
6. Tracking — Log delivery, open, response

---

Implementation Tasks

• Build message template system
• Implement scheduling engine
• Create context assembly layer
• Build prompt builder
• Integrate with Vercel AI SDK (Set up Vercel AI SDK infrastructure #413)
• Add push notification delivery
• Implement message tracking
• Create A/B testing framework
• Add message history view

---

Acceptance Criteria

• Messages send within learned time windows
• Quiet hours never violated
• Contextual messages reference actual events
• Tone matches user preference
• Push notifications open to conversation
• Message history accessible in app
• < 2 second generation time

---

Message Examples by Tone

Warm:

"Hey! Saw you hit 7 days. That's a whole week of choosing you. How are you feeling?"

Direct:

"7 days. What's working?"

Encouraging:

"A whole week! You're building something real here. Keep going!"

---

Related

• Parent: [Epic] Sobers Buddy Implementation #412
• Depends on: Set up Vercel AI SDK infrastructure #413, Design Sobers Buddy character and avatar #414, Build user preference settings #415, Implement pattern learning engine #416, Implement mem0 user memory system #423
• Blocks: Implement slip/relapse supportive response flow #418

---

Part of Sobers v2