Skip to content
CRUXCOAST
All work
Live Case study · 2026

AI Registry

A search and comparison engine for the AI tooling market

The AI tooling market grew faster than anyone's ability to navigate it. AI Registry indexes over 6,000 products behind faceted search, structured comparison, and verified reviews, so a buyer can answer 'which of these actually fits my problem' in minutes rather than afternoons.

6,000+ CANDIDATES RANKED COMPARISON
6,000+
Products indexed
8
Top-level categories
3
Discovery paths
$19
One-time listing fee

The problem

Plenty of listings, almost no comparison

Discovery in the AI tooling market is broken in a specific way. There is no shortage of listings; there is a shortage of comparison. Most directories are a name, a logo, a sentence, and an affiliate link, which tells a buyer nothing about whether a product fits their constraints.

The result is that evaluation happens in browser tabs. Someone shortlists six tools, opens six pricing pages written in six different vocabularies, and tries to hold the differences in their head. The work is real, repetitive, and almost entirely wasted, because the next buyer starts from nothing.

What was built

Five surfaces over one normalised catalogue

  1. 01

    Faceted search over a normalised catalogue

    Every product is normalised onto the same schema: category, pricing model, capability tags, and review data. That normalisation is what makes filtering meaningful rather than decorative, and it is the part that takes the work.

  2. 02

    Structured side-by-side comparison

    Products are compared on the same attributes rather than on their own marketing copy. The comparison view is the reason to come back; the directory is only how you reach it.

  3. 03

    Alternatives as a first-class entry point

    Most real sessions start from a product a buyer already knows. Alternatives pages turn that into a supported path rather than an accident of search.

  4. 04

    Verified reviews tied to purchase

    Reviews are anchored to verified purchases, because an unverified review system in a directory converges on vendor-written content and takes the rest of the catalogue's credibility with it.

  5. 05

    A self-serve listing pipeline

    Vendors submit through a moderated flow with a one-time fee and permanent hosting, handled through Stripe. Supply is a product surface in its own right, not an inbox.

Catalogue shape

Where the depth actually is

Published rather than smoothed over. A buyer can tell in one glance whether their area is well covered, which is worth more than looking uniformly deep everywhere.

  • 2,326

    Productivity

  • 1,488

    Data & Analytics

  • 618

    Business & Finance

  • 452

    Image & Design

  • 432

    Content & Writing

  • 382

    Marketing & Sales

  • 203

    Video & Audio

  • 179

    Development & Code

Decisions

The choices that shaped it

Every one of these traded something away. Those trades are the interesting part of the build.

Category counts are published, not hidden
Showing that Productivity holds 2,326 entries and Development & Code holds 179 is an admission of imbalance. It is also the fastest way for a buyer to calibrate whether the catalogue covers their area, which matters more than looking uniformly deep.
Comparison is the product; the index is the funnel
Directories that optimise purely for listing volume end up competing on SEO against everyone else with a scraper. Building the comparison layer first means the defensible asset is the normalised data, not the page count.
Charging vendors keeps incentives legible
A one-time listing fee rather than affiliate commission means placement is not for sale per click. It is a smaller revenue line and a much easier promise to keep.

Stack

  • Next.js
  • TypeScript
  • PostgreSQL
  • Full-text search
  • Stripe
  • Vercel

Building a catalogue, a marketplace, or a comparison product? The hard part is rarely the interface. Tell us what you are indexing.

hello@cruxcoast.com