WedLoop
An AI concierge that texts with a couple's wedding guests so the couple doesn't have to.
- Next.js
- React
- Tailwind
- Django
- LangGraph
- Claude (Anthropic)
- Twilio
- Stripe
- Firebase Auth
- Cloud Run
- Vercel
- Playwright
Screenshots are from a local build with demo events and guests.
Guests and the concierge
Planning tools
Onboarding
Marketing site
Why I built it
I built WedLoop after a year of friends’ weddings, then my own wedding, experiencing as both a guest and a host what it’s like when seemingly everyone loses your wedding website (or just doesn’t look at it --- ever). I thought this would be the perfect use case for an agent, and that the medium of communication should be where it’s already happening: SMS.
How it works
A couple points WedLoop at their wedding website. It scrapes the site, proposes Q&A pairs as its knowledge base, takes the guest list, and texts every guest. Guests reply with questions and WedLoop answers.
When it isn’t confident, it escalates to the host (the couple or their planner) with a text linking to the question in the portal. The host answers once: the reply goes to the guest through the agent and can be saved as knowledge, so the agent handles that question next time. It also sends reminders for RSVP deadlines, hotel blocks, and travel dates.
What I owned
Everything.
- Product and UI. Onboarding, the host dashboard, the escalation inbox, billing, and the public RSVP page, in Next.js, React, and Tailwind, built against a written design system (onboarding from Figma frames), with a Playwright end-to-end suite.
- The agent. LangGraph flows on Claude, where a classifier decides when a question needs a human and a verifier checks each drafted answer, plus an eval suite that tests answer quality and consistency.
- Backend and infrastructure. Django and Celery on Cloud Run, the Next.js app on Vercel, Twilio, Stripe one-time and subscription billing, and Firebase Auth, with monitoring and alerting.
- Shipping. CI that runs unit tests and agent evals on every change, and automatic frontend and backend deploys.
- Trust. Consent at setup, guest opt-out, delivery safeguards, and usage limits per tier. An agent that texts real people needs to stay quiet when it should.
I also owned every founding engineer’s kryptonite: business development. That meant working a booth at wedding trade shows, cold-emailing and calling wedding planners, and running demos.
What happened
One couple bought it and used it with 200+ guests and 2,000+ messages on one of the most important days of their lives. They loved it.