Mini Report
10 creditsPOST /api/ideate/analyze
Headline subset of the pipeline
- Idea normalization + intake
- Demand pressure check
- Lightweight competitor lookup
- Headline score and quick verdict
The pipeline behind every Unycorn report
Most idea validators are a single LLM call.
We run a 12-stage data pipeline.
Real signals from Product Hunt, Google Trends, deep web search, Reddit, Twitter, GitHub, Wikipedia and live competitor websites — not vibes from a chat model.
5–10 min
analysis time
30+
data points per report
Asking ChatGPT “is my idea good?” is not validation.
Generic AI tools
- Hallucinated competitor names
- Invented market size numbers
- No real-time data — frozen training cutoff
- 30 seconds of confidence, zero evidence
Unycorn
- Real competitor discovery from Product Hunt + curated DB
- Live Google Trends pulls at request time
- Deep-search-verified demand, plus upvoted complaints from Reddit & Twitter
- Multi-LLM cross-validation — disagreements get flagged, not guessed
The pipeline
What happens after you press Validate
Twelve stages. Most of them fetch real evidence. The LLMs come last — and we run multiple of them against each other to catch hallucinations.
- intake
intake.normalize
Idea intake & normalization
Parse the user-submitted idea into category, target user, and core keywords.
- compute
lib/demand-pressure.ts
Demand pressure analysis
Score how much real-world frustration exists around this problem.
- fetch
lib/google-trends.ts
Google Trends pull
Live search-volume curves for the core keywords — momentum, not memory.
- fetch
lib/web-search.ts
Deep web search
Deep search across the open web for organic competitive landscape, ad density, and who actually ranks for the queries that matter.
- fetch
social.signal-mining
Reddit & Twitter signal mining
Scan Reddit threads and Twitter for upvoted complaints, workaround posts, and founder buzz around the problem space.
- fetch
db.competitors · db.competitor_sources
Competitor discovery
Match your idea against our Postgres competitor table, sourced from Product Hunt today (HN/GH next).
- fetch
lib/enrich/competitor.ts
Live homepage fetch + LLM positioning
Fetch each competitor’s site and extract positioning, target users, pricing model, and vertical.
- fetch
lib/competitor-enrichment.ts
Funding / team enrichment cascade
Wikipedia → GitHub → Product Hunt → web search → Apollo. Cached, deduplicated, snake_case keys.
- compute
lib/market-timing.ts
Market timing analysis
Is the wave forming, cresting, or already broken? We score the moment.
- compute
lib/economics-engine.ts
Unit economics modeling
Back-of-the-envelope CAC, LTV, and the price point that has to work for this to be a business.
- synthesize
orchestrate.multi-llm
Multi-LLM orchestration & validation
Multiple LLMs synthesize the report and cross-check each other. If they disagree on a claim, we flag it instead of guessing.
- persist
lib/ideate-scoring.ts
Data aggregation
Roll every signal — demand, social, competitive, economics, timing — into the final structured report and score breakdown.
This is why a real report takes 5–10 minutes.
We're not generating text — we're collecting evidence.
Sources
Where the signal comes from
Every report is grounded in fresh, traceable data. No invented competitors. No made-up market sizes.
Product Hunt
Competitor discovery, votes, comments, Featured signals
Daily updates
Google Trends
Search-volume momentum on the core keywords
Live, per request
Deep web search
Organic competitive landscape, ad density, ranking pages
Live, per request
Upvoted complaints, workaround threads, community pain
Live, per request
Twitter / X
Public sentiment, founder buzz, complaint signals
Live, per request
Hacker News
Technical interest, Show HN launches, Ask HN frustrations
Live, per request
GitHub
Repo signals, stars, activity for technical competitors
Per enrichment, with caching
Wikipedia
Funding history, founding dates for established players
Per enrichment, with caching
Live web fetch
Competitor positioning, pricing, target users — LLM-extracted from homepage HTML
180-day cache, refresh on demand
Quality score
We don't just count competitors. We weight them.
compute_quality_score(id)Postgres function
quality_score = votes + 0.5 × comments + (100 if Featured else 0)
Computed in Postgres via compute_quality_score(id) and refreshed automatically on every Product Hunt ingest. A loud launch beats a dead repo, every time.
Analysis tiers
Two depths of analysis
Same pipeline. Different number of stages run, depending on how much evidence you need.
Full Report
100 creditsPOST /api/ideate/deep
All 12 stages
- Full competitor discovery + enrichment cascade
- Live Google Trends, deep web search, Reddit + Twitter mining
- Market timing + unit economics modeling
- Multi-LLM orchestration & validated final report
Why it takes minutes
Slow on purpose.
A 30-second answer is a 30-second guess. We'd rather take a few minutes and hand you something you can actually defend.
Real fetches, not cached guesses
Every report pulls fresh trend and SERP data at request time. The signal you see is from today, not from someone else’s session last quarter.
LLMs synthesize, then cross-check each other
Multiple models only see evidence we already collected — and they validate the report against each other before you see it. Disagreements get flagged.
180-day cache where it makes sense
Competitor positioning is cached for 180 days because positioning doesn’t change daily. Trends and SERP are always fresh.
Ready to put your idea through the pipeline?
Drop in a startup idea and watch all twelve stages run.