raziel

The description states that raziel is an editorial catalog of AI models, designed to help users make decisions about which models to use for specific tasks. It is described as a Spanish-first product with English as a first-class second language, and it includes features such as searchable model families, detailed model pages, a comparator, a recommendation guide (/choose), and an editorial feed (radar). The project is built using Next.js 16, React 19 Server Components, TypeScript, and Zod for data validation. It is self-reported to be curated by hand with no backend infrastructure currently in place.

The author claims the product avoids AI black boxes and fake confidence scores, instead relying on transparent editorial rules and curated data. The team size is stated as one (Luis Muñoz Martínez). There is no evidence of revenue, customers, or traction beyond the self-reported project description.

• The description states that raziel is an editorial catalog of AI models, with features including:
  • A searchable and filterable catalog of model families and variants.
  • Model detail pages with ratings, benchmarks, architecture, license, local-execution taxonomy, and release timeline.
  • A comparator for 2–3 models, reproducible via URL parameters.
  • A recommendation guide (/choose) that applies transparent editorial rules over curated data.
  • An editorial feed (radar) of model releases linked back to catalog entries.
• It is built with Next.js 16 (App Router), React 19 Server Components, TypeScript strict mode, and Zod.
• The system uses hand-curated JSON, no backend, no scraping, and no database yet.
• The product is described as fully bilingual (es/en), with localization handled via next-intl and copy stored in messages/{es,en}.json.

Not evidenced: What the actual data sources are, how models are selected for inclusion, or whether there is any automated data ingestion.

• The description states that raziel was inspired by the problem of "a wall of leaderboards and marketing copy" that answers a question nobody actually asked.
• It positions itself as "IMDB + Rotten Tomatoes" for AI, with an editorial, human-curated approach.
• The product is described as Spanish-first, with English as a first-class second language.
• The author claims the product avoids AI black boxes and fake confidence scores, instead using transparent editorial rules.
• It is positioned to answer the question: “for this specific job, which model should I try?”
• The evolution of the positioning appears to be from a hackathon project to a curated, data-driven, bilingual AI model catalog, with an emphasis on trust and transparency.

Inferred: The positioning reflects a shift from generic AI tooling toward a more curated, human-in-the-loop approach. However, this is based on self-reporting and not validated by external evidence.

• The description states that the product is aimed at users who want to "make a choice" about which AI model to use for specific tasks.
• It is described as being Spanish-first, with English as a first-class second language, suggesting a bilingual audience in Spanish-speaking regions.
• The target user appears to be someone looking for practical, actionable information rather than scientific rankings or marketing copy.
• The product is described as not for AI black box decision-making, but for human-readable, editorially curated choices.

Not evidenced: No explicit customer personas, usage scenarios, or segmentation beyond the general audience of users needing to choose models.

• The description does not state any pricing model or business model.
• There is no mention of monetization, revenue streams, or paid features.
• The product is described as curated by hand, with no backend infrastructure, suggesting a low-cost, possibly free, approach at this stage.

Inferred: If the project scales, it may move toward a freemium or subscription model, but there is no evidence to support this.

• Built with Next.js 16 (App Router), React 19 Server Components, TypeScript strict mode, and Zod.
• Uses a vertical-slice architecture (app → features → entities → shared).
• Data contract is enforced via Zod, with versioned, hand-curated JSON.
• No backend, no scraping, no database yet.
• Localization handled via next-intl, with copy in messages/{es,en}.json.
• A custom dependency-free theme store instead of next-themes.
• The /choose feature maps user answers to a canonical use-case taxonomy and returns one main recommendation plus up to two alternatives.
• Data-quality tooling includes data validation, dead-code analysis, rating recalibration.
• The project is described as having a written decision log, making its history inspectable like its code.

Inferred: The technical stack suggests a modern, data-first approach with strong emphasis on code quality and maintainability. However, the lack of backend or database implies limited scalability or automation.

• The project is described as being built by one person (Luis Muñoz Martínez).
• It includes 23 model families and 47 variants, curated by hand.
• No evidence of users, customers, revenue, or adoption beyond the self-reported description.
• The product is described as a hackathon submission, suggesting early-stage development.

Not evidenced: No metrics on usage, engagement, retention, or user feedback. No evidence of traction or growth.

• The description states that raziel is inspired by the problem of “a wall of leaderboards and marketing copy”.
• It positions itself as a curated alternative to generic AI rankings, with an editorial approach similar to IMDB + Rotten Tomatoes.
• No specific competitors are named, but it implies a gap in the market for human-curated, practical AI model information.

Inferred: The competitive landscape likely includes AI benchmarking sites and leaderboard platforms, but raziel differentiates itself by focusing on practical use cases, editorial curation, and transparency. However, this is not confirmed by external data.

• The project is described as being built by a single developer, with no backend or database.
• It relies entirely on hand-curation, which may not scale beyond the current scope.
• No evidence of monetization or revenue model.
• The lack of a backend or data infrastructure suggests a high-risk path to growth if it needs to expand beyond current capacity.
• The product is described as a hackathon submission, raising questions about long-term viability and intent.

Inferred: If the project does not scale beyond one person, it may struggle with consistency, data freshness, and user adoption. The lack of monetization or traction also raises concerns about sustainability.

1. What is the plan for scaling beyond hand-curation?
2. How will the product evolve from a hackathon project to a sustainable offering?
3. Are there any plans for monetization or revenue streams?
4. How does the team intend to maintain data quality and freshness as the catalog grows?
5. Is there any interest in partnerships, integrations, or community contributions?
6. What are the long-term goals for localization beyond Spanish and English?