Monitoring Hacker News, Product Hunt and tool registries for product opportunities cost hours a day and stopped whenever the week got busy. We built a pipeline that collects candidates daily, scores them with an LLM against explicit criteria — locale independence, build time, monetisation path — and delivers around fifteen pre-screened ideas as a digest.
Teams that ship small products quickly need a steady inflow of candidate ideas, and the usual way to get one is to read: Hacker News, Product Hunt, new-tool registries, release feeds. Done properly this costs hours every day, and it is exactly the kind of work that degrades when the week gets busy — the reading stops, the pipeline empties, and the next idea comes from whatever happened to be memorable. The signal-to-noise ratio makes it worse: the vast majority of what gets posted is irrelevant to any given team's constraints, so most of the reading time is spent discarding. What was needed was not more sources but a filter that runs whether or not anyone feels like reading that morning.
A scheduled job pulls new entries from Hacker News, Product Hunt and open tool registries, normalising them into a single record shape regardless of source so downstream scoring does not care where an item came from.
Every candidate is scored by an LLM against criteria fixed in advance: locale independence, realistic build time, and a plausible monetisation path. Because the criteria are explicit and versioned, the filter's behaviour can be argued with and tuned rather than trusted blindly.
Survivors are compiled into a short daily digest — roughly fifteen screened ideas — with the reasoning attached, so a five-minute read replaces a two-hour scan. Rejected items stay queryable when a criterion later changes.
The daily research routine collapsed from hours of scanning to a short review of around fifteen pre-screened ideas, each already checked against the criteria that would otherwise have disqualified it later. The pipeline runs unattended every day, which matters more than its per-run quality: the flow of candidates no longer depends on anyone's schedule. It is in production and running on a daily cadence.
Tell us what the process looks like today and we will tell you what can be automated — and what should not be.
LET'S TALK
