Under the hood

The automation & AI pipeline
behind GetReplanted

Built with Claude, Mailchimp, and workflow automation — this is how a solo founder ships a personalised relocation product without an engineering team.

3 live workflows Claude Sonnet + Haiku 5 planned automations Anthropic API Mailchimp MCP Vercel
Live workflows

What's running right now

Two active pipelines — onboarding and specialist consultation — both using Claude as the intelligence layer.

Workflow 1 — User onboarding & AI roadmap generation LIVE
1
LIVE
🌐
Webhook trigger
User submits onboarding form
2
🔀
Parse profile
Extract 6 answer fields
3
⚙️
Branch logic
Degree / family / German level
4
AI
🤖
Claude Sonnet
Generate personalised summary
5
🗺️
Build roadmap
6–9 steps injected by profile
6
💾
Persist state
localStorage + DB
7
✅
Return JSON
Roadmap delivered to client
Workflow 2 — Speak with a specialist & Mailchimp routing LIVE
1
LIVE
📋
Specialist form
User requests a callback
2
📥
Capture & validate
Email + visa type fields
3
📬
Mailchimp
Add + tag by visa type
4
AI
✍️
Claude Haiku
Write personalised intro email
5
📧
Send to user
Specialist reaches out in 24h
6
📊
Log request
Visa type + source tracked
Workflow 3 — GitHub → Vercel continuous deployment LIVE
1
✏️
Code change
App.jsx or index.html updated
2
LIVE
🐙
GitHub commit
Push to main branch
3
🔗
GitHub webhook
Triggers Vercel build
4
📦
Vite build
React app compiled
5
🌍
Vercel deploy
Live on app.getreplanted.life
6
✅
Production live
Zero downtime, ~30s deploy
API payloads

The actual JSON

Real request and response objects from the live pipeline — every field drives a specific decision in the roadmap.

claude_api_request.json
Node 4 — AI
{
"model": "claude-sonnet-4-20250514",
"max_tokens": 400,
"system": "GetReplanted guide.
Build an honest 3-sentence
summary. Name one key
challenge directly.",
"messages": [{
"role": "user",
"content": {
"country": "india", // → APS cert step
"situation": "job_hunting", // → Chancenkarte
"timeline": "urgent", // → embassy flag
"degree": "unsure", // → eval step
"family": "partner", // → family visa
"german": "basic" // → language step
}
}]
}
roadmap_state.json
Node 5 — Output
{
"summary": "You're going for the
Chancenkarte from India.
Your APS cert is mandatory
— start it first.",
"steps": [
{ "id": 0, "title": "Apply for Chancenkarte" },
{ "id": 1, "title": "Check degree recognition" },
{ "id": 2, "title": "Open blocked account" },
{ "id": 3, "title": "Get health insurance" },
{ "id": 4, "title": "Apply for family visa" },
{ "id": 5, "title": "Find temporary housing" },
{ "id": 6, "title": "Register your address" },
{ "id": 7, "title": "Start German lessons" }
],
"step": 0, "done":
}
By the numbers

Built fast. Ships faster.

3+
Live workflows
6+
Onboarding inputs
9+
Max roadmap steps
5+
Planned automations
0€
Engineering cost
Roadmap

What we're building next

Five automation pipelines in design — each one removing a manual bottleneck or unlocking a capability that doesn't exist yet.

🔑
SEO content refresh pipeline
Ahrefs MCP + Vercel MCP
1
Weekly scheduled trigger
2
Ahrefs MCP → fetch keyword rankings for all 8 guide URLs
3
Claude Sonnet → identify gaps vs. top 3 ranking pages
4
Generate updated sections with current 2026 data
5
GitHub MCP → open PR · Vercel MCP → deploy on merge
Keeps all guides ranking without manual effort
🧠
AI step advisor (in-roadmap Q&A)
n8n webhook
1
User asks a question on any roadmap step
2
Inject user profile + step context into prompt
3
Claude Sonnet → answer in max 3 sentences + one action
4
Log Q&A to Airtable for product learning
5
Flag recurring questions → feed back into static guides
Re-enables AI layer without always-on API cost
📈
Affiliate performance tracker
Claude Haiku
1
Daily trigger → pull click data from affiliate dashboards
2
Normalise across Expatrio, Fintiba, Feather, N26, Wise
3
Claude Haiku → flag underperforming placements
4
Auto-update AFF config if better alternative detected
5
Weekly Slack digest → revenue + click summary
Turns affiliate revenue from passive to actively optimised
🌍
New country expansion pipeline
Research MCP
1
Trigger → new country added to onboarding config
2
Research MCP → pull embassy wait times + visa requirements
3
Claude Sonnet → generate country-specific tips for all 8 guide sections
4
Draft roadmap variants for new country profile
5
Human review step → approve before going live
Scales from 6 countries to 20+ without manual research
📬
Drip sequence personalisation
Mailchimp MCP
1
Trigger → user completes a roadmap step
2
Pull profile → identify next 2 steps in their sequence
3
Claude Haiku → write personalised nudge email
4
Mailchimp MCP → send at optimal time by timezone
5
Track open + click → feed back into model tuning
Moves from broadcast to 1:1 step-aware nudges