Verdict

Inspiration

We were building our own startup and genuinely had no idea if it was good.

Not in a humble way — we actually couldn't tell. We'd sink hours into market research: 30 browser tabs open, outdated Statista reports, trying to figure out if a competitor had shut down or was still active, copy-pasting TAM numbers that felt completely made up. Hours gone. And we'd still end up in the same place — uncertain, second-guessing, not knowing if we should keep going or just stop.

That's the frustration that built Verdict. We wanted something that would just tell us the truth — the way a sharp investor would in the first 10 minutes of a pitch. Live market data. Real competitors. An honest score. Not vibes, not encouragement — a verdict.

What it does

Verdict validates startup ideas in 60 seconds using live web intelligence.

You paste your idea, answer 6 quick questions about your founder profile (stage, budget, technical skill, geography), and get a full report: real market sizing (TAM/SAM/SOM with sources), live competitor intel, a market entry score, a Go/No-Go viability verdict, and a Devil's Advocate section — 3 companies that tried something similar and exactly why they failed.

Every section is grounded in live web sources via Perplexity's Sonar Pro API, not stale training data. And because the same idea looks totally different depending on who is building it, the entire scoring system personalizes to your founder profile.

How we built it

• Frontend: Next.js 15 App Router + Tailwind CSS + Framer Motion
• AI layer: 6 parallel Perplexity Sonar Pro calls — each scoped to a single section (market sizing, competitor intel, entry barriers, viability verdict, failure cases, idea snapshot) so nothing blocks anything else
• Scoring: Custom rubrics per section with dynamic color coding — founders read the signal in half a second, no parsing required
• Smart re-validation: When a founder updates their profile, we skip the 4 static sections and only re-run the 2 founder-sensitive calls — ~60% faster
• Export: Native print-to-PDF with a custom print stylesheet. No html2canvas hacks. Pixel-perfect.

Challenges we ran into

Getting honest scores. LLMs are optimistic by default — they want to encourage you. Getting Perplexity to tell a founder their idea is a No-Go when it genuinely is took way more prompt engineering than we expected. Explicit rubrics, calibrated thresholds, and a lot of iteration.

Reliable structured output. Sonar Pro returns markdown-fenced JSON with inline citation markers ([1], [2]) baked mid-string. Any one of our 6 parallel calls breaking would kill the whole report. We built a parsing pipeline that strips fences, citation markers, and leading prose before touching JSON.parse().

Founder context changing everything. The same idea scores differently based on who's building it. Wiring the founder profile into every prompt — not just as a footnote but as a core scoring input — was a design challenge as much as a technical one.

Accomplishments that we're proud of

• A full investor-grade report in under 60 seconds, sourced from the live web
• The Devil's Advocate section — real companies, real failures, the pattern to avoid. No other tool does this.
• A scoring system that asks who's building before it scores what's being built
• A UI that makes a dense, multi-section report feel scannable and trustworthy at a glance — not overwhelming

What we learned

Live search is a fundamentally different product than an LLM answering from memory. The citations make every claim checkable — and that changes how much a founder trusts the output. We also learned that making AI be honest is harder than making it seem impressive. And honestly — building Verdict taught us more about our own ideas than any market research we'd done before.

What's next for Verdict

We're not shelving this. After the hackathon we're watching what judges and early users actually respond to, then deciding what to double down on. The immediate roadmap: auth and payments (Supabase + Stripe) to turn it into a real product, report history so founders can track how their thinking evolves, and team sharing so co-founders can review and react async. If the signal is there — we're building.

Built With

• next.js
• perplexity
• react
• tailwind
• typescript
• vercel