Ryan Chin · Founder, Rainy Day Ventures

I build AI products end to end.

Products that catch every inquiry. From the customer problem and the spec, through the architecture and the code, to the pricing page and the first paying user. I started and built four of them at Rainy Day Ventures: two live, two on the way.

Portrait of Ryan Chin
2 products liveweb, iOS, Android
2 shipping soonBetterfy, Yesroom
Hi, I'm Ryan
4products founded and built
6,800+commits across the four codebases
5,000+automated test files
3platforms shipped: web, iOS, Android

The ventures

Each one started as a problem I wanted solved, and I took each one from idea to launch. Click Try it on any card to play with a small demo of the idea.

Coming soonAI agents · SaaS

Betterfy

The website that runs the business behind it.

An AI platform for small-business owners. It builds an industry-specific website from a few questions, then acts as an AI operations team: it catches every inquiry, drafts the follow-ups, blog posts and support replies, and tells the owner who to call today. The AI does the work; the owner approves it.

Next.jsTypeScriptSupabasePostgres RLSStripe ConnectPostHogOpenRouterVercel
betterfy.ai →
Rebuilt from betterfy.ai · try it
LiveLLM evaluation

PromptJury

Put your prompt on trial.

An evaluation tool for teams shipping LLM features. A guided six-step wizard proposes a grading guide, has a panel of models draft competing prompts, writes test cases, runs every candidate model, and has a blind jury grade the answers. The verdict: which prompt and model to ship, how sure you can be, and what each costs.

Next.jsTypeScriptSupabaseInngestStripeZodPlaywright
promptjury.com →
Rebuilt from promptjury.com · pick a case
Swearing in the models…

Live · iOS & AndroidConsumer mobile

Growbook

Capture, cherish, celebrate every chapter of your child's life.

A private digital baby book. Parents save photos, videos and milestones and share them securely with family, from first steps to graduation day. Available on the App Store and Google Play.

Ruby on RailsPostgreSQLRedisCloudflare Images & StreamAWS RekognitioniOSAndroid
getgrowbook.com →
Milestone saved: first steps
Growbook app showing a child's timeline
Coming soonFintech · iOS

Yesroom

Know what's left before you spend.

Household budgeting that answers at the moment of purchase. Banks see completed transactions; Yesroom sees intent: the shopping list, the calendar event, the store you just walked into. It tells you "still fits" or shows what is left, while the decision can still change.

Expo / React NativeTypeScriptSupabasefast-checkTurborepo
yesroom.com · launching soon
Groceries · November
$238 left
Still fits
ILLUSTRATIVE DEMO · DRAG THE SLIDER

What I bring to an AI product team

These skills usually belong to three different people. On every product above, I did all three.

Hiring for…

Pick a role to see the evidence that matters most for it.

Product

Deciding what to build, and proving it works.

  • Discovery to launch: personas, positioning, pricing and go-to-market for each venture.
  • Specs that ship: milestone plans with acceptance criteria and written decision records.
  • Honest claims: marketing promises tied to features that verifiably run in production.
  • Experimentation: A/B testing and product analytics built in from the start.

AI

Making LLMs reliable enough to put in front of customers.

  • Evals: built a whole product for them. Blind LLM-as-judge grading with statistical confidence.
  • Model routing: one gateway across Claude and open-weight models, with cost budgets.
  • Structured output: every model response validated against a schema before it is used.
  • Human in the loop: AI drafts, people approve, and numbers come from code, not models.

Engineering

Full stack, from database policy to app store.

  • Web: Next.js, TypeScript, React; Ruby on Rails APIs.
  • Mobile: Expo / React Native; a consumer app live on iOS and Android.
  • Data: Postgres and Supabase, multi-tenant row-level security, background jobs.
  • Quality: test-first, with unit, end-to-end, property-based and mutation testing.

How I build with AI

I work with AI coding agents every day. These four rules make that fast without making it fragile. Tap one to open it.

Every product starts as a written spec with milestones and acceptance criteria. Agents build against it one milestone at a time, and every non-obvious choice gets a short decision record.

LLMs classify, draft and parse into typed structures. Anything a customer relies on, such as a dollar figure or a score, is calculated by deterministic, tested code.

A feature isn't finished until its tests pass, and I break the code on purpose to confirm the tests actually catch it. Over 5,000 test files across the four products.

Prompt and model choices are decided by blind, graded evals on real cases, with the cost of each option on the table. That's why PromptJury exists.

Writing

Notes from building AI products in the open

What I'm learning about shipping with AI, product decisions, and running small ventures. On Substack.