or a few years now I've taken on freelance dev projects on the side. After
enough of them, I started noticing something: different clients, wildly
different projects on the surface, but the same underlying need showing up
again and again underneath.
A client wants a booking calendar. Another wants a booking calendar with a
different color scheme. A third wants "basically what the first two wanted,
but for dentists." Same shape, different wrapping.
Once you see that pattern, the obvious move is to stop rebuilding the same
thing from scratch every time — extract the common piece, reuse it, ship
faster, bid lower and still make more per hour. I did that manually for a
while: skim job boards, eyeball which requests rhymed, build the reusable
version. It worked, but it didn't scale past my own attention span.
Why "what should I build next" is usually answered by guessing
Every "what should I build" thread I've seen follows the same shape: someone
throws out an idea, a few people say "cool," nobody has any evidence either
way. Even asking an LLM for ideas hits the same wall — it's fast, but the
suggestions are generic pattern-matching over public discourse, not signal
from what people are actually paying for right now.
Search volume isn't it either. More people search "how to lose weight" than
search for any SaaS product on earth, and that tells you nothing about
whether they'll pay you for anything.
The one thing that does correlate with real willingness to pay: people
who already opened their wallet for it. Not a survey answer, an actual paid
task on a freelance platform.
So I built the thing I was doing manually, as software
Idea Miner pulls real paid tasks from freelance/outsourcing platforms
(Freelancer.com to start), clusters the recurring ones with LLM-assisted
grouping (batched per cluster, not per task — cost matters), and tracks for
each recurring pattern: how often it gets posted, the median budget, and an
opportunity score that weighs demand against how much competition already
exists for it.
Right now that's 2,686 distilled demand patterns, traced back to
25,900 real paid tasks. Every number links back to the actual task IDs
it's built from — nothing here is a vibe, it's an aggregation of receipts.
One thing worth being upfront about: a high opportunity score describes one
specific cluster, not "the market" in general — most of what shows up is
niche and un-sexy (site-verification gigs, SEO cleanup, compliance
paperwork), not the next unicorn category. It just tells you people are
already paying for it, repeatedly, right now.
Stack, for the curious
Django + Postgres, LLM-assisted clustering, deployed on a single Ubuntu box
behind Gunicorn/Nginx. Deliberately boring — no Celery, no vector DB in the
early phase, nothing that isn't earning its complexity yet.
Where I'm at
Pre-revenue, figuring out whether the ranking board itself is useful enough
that people come back to it weekly, or whether it needs to be paired with
something more active (a community, alerts, validation reports) to actually
change what people build.
If you're the type who has the skills but not the direction, it's at
https://idea-miner.com/en/. Curious what you think — and especially curious
whether "real paid demand" actually changes anyone's decision, or if it's
just interesting data to look at once and forget.
