Skip to main content
itschijong//v.04
let's talk
Tooling / AutomationActive

App Idea Factory

Three-pipeline automation for surfacing, scoring, and specifying app ideas

Solo build·2025·Active
SupabaseVercelClaude APIThree decoupled pipelines

Context

I generate more product ideas than I can act on. I wanted a system that would surface, score, and — on demand — deeply specify ideas, without me sitting at a keyboard for it.

What I built

A pipeline architecture with three decoupled stages running on Supabase + Vercel.

How it works

App Idea Factory — Pipeline Architecture

Stage 01

Scraper

Scheduled signals from Reddit, IndieHackers, etc.

Stage 02

Curator / Ranker

Dedupes, filters noise, scores against criteria.

Stage 03

Spec Generator

On-demand: generates full implementation spec via Claude API.

Stage 1 — Scraper. Pulls signals from sources of demand (Reddit, IndieHackers, etc.) on a schedule.

Stage 2 — Curator / Ranker. Filters noise, dedupes, and scores ideas against criteria I define (market size proxy, technical feasibility, alignment with what I'd want to build).

Stage 3 — On-demand Spec Generator. When I select an idea, this stage generates a full implementation spec ready to hand off to Claude Code.

The decoupling matters: each stage runs on its own schedule, has its own failure mode, and can be reasoned about independently. A coupled monolith would have been faster to ship but would not survive contact with real-world data quality.

Outcomes

Personal use, ongoing. The interesting general lesson is that AI-augmented automation works best when you treat the LLM as one component in a pipeline with clear boundaries — not as a magic everything-machine.