Verdiqt

Inspiration

AI made building software cheap. It did not make building the right software cheap. People now ship a SaaS product in a weekend, and find out at launch that nobody wanted it.

Verdiqt moves that painful discovery to before you write the first line of code. You put your idea on trial, and a court decides whether it deserves to live.

What it does

You file a case: your SaaS idea in a sentence, or a link to a public GitHub repo you already built. Then the courtroom goes to work, and you watch every step happen live:

The case file. The court reads your idea and writes it up: who it's for, what problem it solves, what to search for.

The investigation. Real research across five real communities : Hacker News, Reddit, GitHub, Stack Overflow, and Product Hunt. Every finding appears on screen with a link to its source, and every source reports honestly: gathered, failed, or empty. If the first search comes back thin, the court automatically widens its search.

The deliberation. Six AI judges each take one question: "Is the pain real? Is anyone actively looking for this? Who already exists? Who would pay? Can you reach these people? Can you build it fast enough?" Each judge delivers their score the moment they finish, you literally watch the bench board fill in, one verdict at a time.

The ruling. A presiding judge reads the entire case file and rules: BUILD, PIVOT, or KILL —> with a score out of 100, the reasoning behind every number, and exactly one recommended next step (a real, cheap test you can run this week).

A full trial takes about a minute. And if you refine your idea and try again, the court remembers: the evidence you marked as important and the previous ruling follow your case, so every iteration builds on the last.

How we built it

Zero cash out of pocket. The whole product runs on free tiers and hackathon credits: one small server runs the website and the research pipeline, the AI judges run on free model APIs, and when one provider's daily quota runs dry, the court automatically switches to a backup provider mid-trial and keeps ruling. Constraints became design: the court paces itself under rate limits, shows its deliberation while it works, and never wastes a token.

The model argues; the code rules. Anti-hallucination isn't a prompt, it's enforced in code. Judges may only cite evidence the court actually gathered: invented citations are stripped automatically, and a high score that loses its citations gets capped. When public evidence is thin, the judge is allowed to use domain judgment but the ruling must say so, out loud, in the rationale. Uncertainty is never dressed up as certainty.

Privacy without accounts. No sign-up. Your browser gets an unguessable key; only its fingerprint is stored. Your cases are yours, anyone else asking for them gets an identical "not found.", don't worry.

Challenges we ran into

Running a multi-stage AI pipeline on free tiers meant treating every token and every request as scarce... the wait-and-retry pacing, the provider failover, and the one-at-a-time trial queue all exist because of it. And WebMCP is new enough that the tool contract itself was a design problem: we learned that the workflow belongs inside the tool descriptions, so agents teach themselves the journey.

Accomplishments that we're proud of

A complete, evidence-cited validation trial in about a minute. A deliberation you can actually watch. A surface that serves two audiences at once... humans get a courtroom, agents get a toolbox, and each can see the other working.

What we learned

Never think you have enough time to build something.

What's next for Verdiqt

GitHub sign-in to put your whole repo graveyard on trial at once, re-trials over time to watch a verdict change as markets move, and Reddit's expanded API access for even deeper community evidence.

Built With

• claude
• codex
• groq
• next.js
• prisma
• render