One catalog. Multiple discovery dimensions.
Universal Entity Registry
Stable IDs for every offering.
Hierarchical Taxonomy
Governed domains & subcategories.
Dynamic Attribute System
Core model + type-specific schemas.
Relationship Graph
Integrations, dependencies, alternatives.
AI Classification Engine
Suggestions with evidence & confidence.
Search & Comparison
Keyword, facets, synonyms, semantic.
Review & Trust
Reviews on canonical records only.
Catalog Governance
Versioned, logged, reversible.
Universal entity types
Every listing has one primary type and may carry additional types. One company can own many offerings; one offering can have many capabilities.
| Type | Examples |
|---|---|
| Software Application | VS Code, Photoshop, VLC |
| SaaS Product | Notion, Slack, Jira |
| Developer Tool | Postman, Docker Desktop |
| API / API Service | Stripe API, Twilio API |
| Library / Package | NumPy, Lodash, React |
| Framework | Django, Spring Boot, Next.js |
| Platform | AWS, Azure, Vercel |
| CLI Tool | kubectl, Terraform CLI |
| SDK | AWS SDK, Firebase SDK |
| AI Model | Llama, Gemini model families |
| Database / Data Engine | PostgreSQL, Redis, MongoDB |
| Infrastructure Component | Kubernetes, Envoy Proxy |
| …plus Website, Plugin, AI Agent, Runtime, Template, Digital Service | |
Shared core + specialized attributes
A SaaS product needs subscription pricing. A Python library needs versions and dependencies. An API needs auth, endpoints and rate limits. One shared core model, plus type-specific attribute schemas — aligned with Schema.org SoftwareApplication and SoftwareSourceCode.
| Common fields | Specialized fields |
|---|---|
| Name, description, website | API authentication & limits |
| Publisher / owner | Package manager & dependencies |
| Primary type | Programming languages |
| Categories & tags | Deployment models |
| License / commercial model | SDK support |
| Last verified | Cloud & infrastructure support |
| Source evidence | Security advisories |
| Reviews & ratings | Version & release data |
PostgreSQL logical data model
One catalog_items row per unique offering, with every other dimension in its own table. Libraries and packages also store a Package-URL (PURL) — a standard identifier across npm, PyPI, Maven and more — to prevent duplicates.
Conceptual model — not the complete production schema.
-- One canonical record per offering CREATE TABLE catalog_items ( id UUID PRIMARY KEY, slug TEXT UNIQUE NOT NULL, -- canonical URL name TEXT NOT NULL, primary_type TEXT REFERENCES types(id), org_id UUID REFERENCES organizations(id), website TEXT, purl TEXT UNIQUE, -- pkg:npm/react attributes JSONB DEFAULT '{}', -- type-specific schema status TEXT DEFAULT 'draft', last_verified TIMESTAMPTZ ); -- Stable category IDs; names may change CREATE TABLE categories ( id TEXT PRIMARY KEY, -- 'cloud.kubernetes' pref_label TEXT NOT NULL, parent_id TEXT REFERENCES categories(id), status TEXT DEFAULT 'proposed' ); -- Many-to-many: one item, many categories CREATE TABLE item_categories ( item_id UUID REFERENCES catalog_items(id), category_id TEXT REFERENCES categories(id), confidence NUMERIC(3,2), -- AI suggestion score PRIMARY KEY (item_id, category_id) ); -- Typed edges of the technology graph CREATE TABLE item_relationships ( from_item UUID REFERENCES catalog_items(id), rel_type TEXT CHECK (rel_type IN ('part-of','integrates-with', 'depends-on','alternative-to','competes-with', 'built-with','sdk-for')), to_item UUID REFERENCES catalog_items(id), PRIMARY KEY (from_item, rel_type, to_item) );
The admin panel behind the catalog
Category changes are versioned, logged, reversible and tested before publication. AI may propose new categories; each must pass duplicate, synonym and scope checks.
Avoid the faceted-navigation trap
A large taxonomy can accidentally generate millions of near-identical filter URLs that waste crawl budget and slow discovery of important content.
- One canonical URL per catalog item
- Index category pages only when they add real value
- Keep arbitrary filter combinations out of the index
- Redirects whenever categories merge or rename
- Structured data appropriate to each listing type
- No thin auto-generated product × type × feature pages
From MVP to 10 million+
| Phase | Listings | Categories |
|---|---|---|
| MVP | 1,000 | 100 |
| Phase 2 | 10,000 | 300 |
| Phase 3 | 100,000 | 1,000 |
| Phase 4 | 1 million | 3,000 |
| Long-term | 10 million+ | As required |
Proposed architectural targets, not current catalog size. Capacity must be validated through load testing — and accuracy, discoverability and real use matter more than raw listing counts.
See the model in action
Explore the catalog with adaptive filters, or open a canonical listing to see all five dimensions.