Hostshare

What the engagement looked like.

Hostshare arrived with an idea, a founder, and no product. The brief: take a novel real-estate primitive — hosts trading nights in each other’s short-term rentals instead of paying cash — and ship something beta-worthy in under a quarter. It has since grown into a full production platform that I own end to end.

The problem on arrival.

No design system, no engineering patterns, no product surface — just a business model and a waitlist building in the background.

It could never be just a UI. A trustworthy marketplace where property and money change hands needs payments, identity verification, real-time messaging, calendar sync with the platforms hosts already use, a payout ledger, and the background machinery to keep all of it honest — built, and kept running, by essentially one person.

←→

How it came together.

6 phases, each one building on what the last one shipped.

1. Weeks 0–2
   System first

   Typography, token primitives, and a component catalogue scoped to likely product needs — before touching a single screen.

2. Weeks 2–6
   Flows as wireframes

   End-to-end host and guest journeys as low-fidelity wires, validated against the founder weekly and iterated.

3. Weeks 6–16
   Vertical slices to beta

   Each slice shipped a working screen on real Hono + Postgres endpoints, not mocked data — then a pass on onboarding, empty states, responsive and motion. Flipped the switch for the waitlist at week 16.

4. Post-beta
   From app to platform

   Grew the single app into a Bun/Turborepo monorepo — the API split onto Hono, a Drizzle/Neon data layer, and a three-tier GitHub Actions pipeline that gives every feature branch a full preview environment with its own database.

5. Then
   The money & trust layer

   Stripe payments, Connect payouts and a credits ledger; Persona identity checks; and the webhooks plus reconciliation jobs that keep every balance and payout honest.

6. Ongoing
   Real-time, sync & AI

   Two-way Airbnb calendar sync, real-time messaging over Ably, a notification pipeline across email and SMS, 40+ background jobs, and Claude wired into message drafting, review summaries, translation, and demand-scored blockout suggestions.

The platform behind it.

1. One monorepo, the right runtime per job

   A Bun + Turborepo workspace of 8 apps and 9 shared packages: a Next.js web app, Hono APIs on Vercel Functions, a Lambda@Edge CDN for images and calendar feeds, a Trigger.dev worker for durable jobs, and Python scrapers — all sharing one typed framework and Drizzle/Postgres data layer.

   BunTurborepoTypeScriptNext.jsReactHonoTailwind CSS
2. A three-tier deploy pipeline

   GitHub Actions orchestrates every release. Each feature branch gets a full preview environment with its own Neon database branch; production is pinned to an immutable release tag and staged behind a DNS swap, health-checked with one-command rollback. Env is driven from AWS Secrets Manager, never touching production by accident.

   GitHub ActionsVercelNeNeonAWSAWS
3. An event, job & notification pipeline

   40+ Trigger.dev tasks — scheduled (exchange-rate refresh, nightly cleanup, payout reconciliation, calendar purge, offer expiry, dashboard snapshots) and event-driven — fan out into a notification layer that routes email, SMS and push through Customer.io and Twilio, mirrored to analytics and conversion APIs, with webhook retries.

   TrTrigger.devRedisCioCustomer.ioTwTwilioPostHog
4. Real-time messaging & Airbnb sync

   Host-to-guest messaging runs in real time over Ably. Calendars two-way sync with the platforms hosts already run — iCal feeds served from the edge plus Hospitable reconciliation — and Python enrichers pull listing data from Airbnb and VRBO.

   AbAblyAirbnbHsHospitableMapbox
5. Payments, payouts & trust

   Stripe runs subscriptions, Connect payouts and a credits-and-wallet ledger; Persona handles identity verification; and a web of webhooks and reconciliation jobs keeps every balance, payout and booking honest.

   StripePePersonaPostgreSQLDrizzle
6. In-product AI, tastefully applied

   Claude powers message drafting and enhancement, review summaries, translation, offer-instruction compiling, and the demand-scored “suggest my blockouts” calendar — each wired as a job, not a chat box.

   Claude
7. Built to be run by agents

   The repo itself is engineered for AI: a shared CONCEPTS.md domain vocabulary, a skills and slash-command system, MCP scaffolding, and a documentation system that compounds learnings — so features ship faster the longer the project runs. Observability is OpenTelemetry tracing, structured logs, and PostHog, on compute region-pinned next to the database.

   ClaudeMCPLinearOpenTelemetryPostHog

What shipped.

Public beta shipped on schedule with 1,000 accounts drawn from a 10,000-deep waitlist — on a platform that now spans a typed Bun/Turborepo monorepo, a three-tier deploy pipeline, 40+ background and scheduled jobs, real-time messaging, Airbnb calendar sync, Stripe payments and payouts, and Claude-powered in-product AI. The system laid down in the first weeks has absorbed every one of those without a rewrite.