How it works
It starts with your builder profile. We embed it into a vector and compare it against every idea using cosine similarity. But matching is only half the story — every idea passes a 6-stage analysis pipeline before it reaches your feed.
Matching
You don't browse a list hoping something looks good. You tell us what you're good at, and we do the work.
Add your skills, technical background, industries you know, and what kind of product you want to build. This takes about two minutes.
Your profile gets converted into a high-dimensional vector — a numerical fingerprint of your builder identity. Every idea in the library has one too.
We compare your vector against every idea's vector using cosine similarity. The closer the match, the higher your fit score. Your For You feed shows the top results, ranked by fit.
The pipeline
About 80% of ideas get filtered out along the way. What survives has been scored, stress-tested, researched, and planned.
We pull raw signal from Reddit startup communities and Y Combinator company data. The system looks for real problems people describe — complaints, workarounds, feature requests, pain points. No brainstorming. No prompts asking an AI to make something up.
Produces
A raw idea with source links, community context, and demand signals.
Filters
Duplicate detection and basic quality gates remove spam, joke posts, and ideas that are just product requests for existing tools.
A rules-based scoring pass evaluates each idea on specificity, market size signals, monetization potential, and community engagement. This is fast and cheap — it's meant to cut the obvious noise before the expensive AI stages.
Produces
A pass-1 score that ranks ideas by surface-level quality.
Filters
Ideas below the heuristic threshold get dropped. This removes about half the intake.
Claude reads the full idea context and evaluates it in depth. It assigns an opportunity score (0-100), a verdict (Strong Opportunity, Worth Exploring, Crowded But Possible, Already Solved, or Too Niche), and writes a refined problem statement, target customer, and proposed solution.
Produces
Opportunity score, verdict, refined problem/solution, target customer description.
Filters
Ideas that score below the analysis threshold or get a negative verdict are removed.
A second AI pass tries to kill the idea. It looks for fatal flaws (things that can't be fixed), fixable flaws (things that are hard but possible), hidden risks the first analysis missed, and historical failure patterns — cases where similar ideas have been tried and failed, and why.
Produces
Devil's advocate score, survival verdict (Resilient, Vulnerable, Fragile, or Dead on Arrival), and a detailed breakdown of flaws and risks.
Filters
Ideas rated Dead on Arrival get removed. Fragile ideas are flagged but kept if other scores are strong.
Perplexity runs real-time web research to find actual competitors. Not AI-generated guesses — real companies with real websites. For each competitor, the system collects pricing, strengths, weaknesses, and market position. It also gathers market signals: TAM estimates, trend direction, G2/Capterra sentiment, funding activity, and regulatory notes.
Produces
3-5 competitor profiles, market signal summary, competition level rating (Blue Ocean, Underserved, Competitive, or Saturated).
Filters
Ideas in saturated markets with no clear differentiation angle get flagged. They stay in the library but are ranked lower.
The final stage produces a one-page business plan for each surviving idea. It includes an elevator pitch, validation strategy (how to test before writing code), MVP scope with tech stack and timeline, target market with personas, go-to-market strategy, competitive positioning, unit economics (CAC by channel, LTV, payback period, gross margin, break-even customer count), retention analysis, pivot pathways, key metrics, risk mitigation, an 8-week action plan, and cost estimates.
Produces
A complete business plan you can actually use to start building.
Filters
None — this is the final stage. Ideas that reach here have passed everything else.
Out of 40,000+ ideas ingested, roughly 20% make it through all six stages. The rest get cut at various points — bad signal, weak opportunity, fatal flaws, saturated market, or not enough differentiation. The ones that survive get matched to builders like you.
Set up a builder profile and we'll rank every surviving idea by how well it matches your skills, experience, and goals.