Ideastorm

Overview

Ideastorm is a SaaS platform for builders and founders who want a steady stream of validated startup ideas. It aggregates idea-rich content, enriches it with AI, and exposes everything through a searchable dashboard instead of forcing users to scrape Reddit threads by hand.

My role & scope

I acted as full stack developer and technical architect, taking the project from initial scaffolding to a production-ready MVP in roughly 6–8 weeks. That included the Next.js app, the database schema, and the AI ingestion pipeline.

Under the hood

• Next.js (app router) for the marketing site and authenticated dashboard, mixing server components for data fetching with client components for rich UI and animation.
• PostgreSQL + Drizzle ORM to model ideas, scores and metadata as typed schemas with jsonb fields for structured AI output.
• NextAuth + Drizzle adapter to secure the dashboard and keep sessions tied to the same schema the analytics uses.
• OpenRouter LLM API in a separate scraper project that turns raw community posts into structured business intelligence (scores, SWOT, suggested tech stack, MVP features, and more).

Interesting challenges

The main challenge was getting from messy, unstructured posts to a schema the product can trust. I designed a constrained JSON output format, defensive parsing around the LLM responses, and a schema in Drizzle that maps 1:1 to the analysis result. That lets the dashboard slice ideas by difficulty, market potential or niche without bolting on extra transformation layers.

What I’d improve next

I’d add stronger runtime validation (e.g. Zod) around AI responses in the ingestion pipeline, plus integration tests that cover the whole flow: scrape → analyze → insert → dashboard query. On the product side, richer filtering and saved-idea features would turn the dataset into a more personal discovery tool.