A tool that mines user complaints on social media to find SaaS niches
A tool that mines user complaints on social media to find SaaS niches
Product ideas usually come out of someone’s head, and then six months go into finding out whether anyone needed them. I wanted to approach it from the other end: first find what people complain about out loud and repeatedly, and only then think about a solution. The raw material is public — comments on VC.ru, Habr and Reddit.
The pipeline
Scrapers for the three sources collect comments, then a language model clusters them into “pains”: recurring complaints condensed into a single statement. Product ideas are generated per cluster, each checked for demand against Google Trends and search metrics, deduplicated, and passed into a revenue-potential calculator.
One detail measurably improved the output: ideas are validated not by one prompt but by three distinct roles — a tech lead looks at feasibility, a product manager at value, a marketer at channel. A single “evaluate this idea” prompt returns an evenly optimistic review of anything; three roles with different criteria start disagreeing, and the disagreement is more useful than the verdict.
Technically
Node with Express on the backend, Next.js on the front, eight Postgres migrations. Long operations — scraping and clustering — report progress over SSE; otherwise the UI shows an endless spinner and nobody can tell whether anything is happening.
The project was built out of personal interest and without git, so dating comes from file timestamps: the core appeared in late November 2025, with another pass in January 2026. The tool works, but it is a tool for myself rather than a product — its main value turned out to be showing how few ideas survive a demand check.