getfrp
ARCHITECTURE

AI Agent × Structured Data × Engineering Delivery

The base of f1frp is not just a database website. It is a three-layer stacked system: the top AI Agent layer consumes the two below it, the middle data layer is a knowledge graph of six interlinked libraries, and the bottom engineering layer turns answers into artifacts engineers can ship with.

THREE-LAYER

Each layer owns its role and feeds the others

AI AGENT
AI Agent layer

Dual-model routing: Gemini 2.5 Flash and Claude Haiku via OpenRouter. pgvector RAG retrieves embeddings across the six libraries, citation-first enforces source attribution. Question → semantic retrieval → sourced answer, never fabricated.

STACK · @ai-sdk/google · @ai-sdk/anthropic · pgvector · OpenRouter fallback

DATA
Structured data layer

Six interlinked tables: materials (4,341) · formulas (102) · standards (95) · papers (698+) · patents (1,473+) · supplier_listings (67). Every parameter traces to its source, every slug deep-linkable via hash anchor.

STACK · Neon Postgres · Drizzle ORM · pgvector · Zod schema

DELIVER
Engineering delivery layer

RFQ form → Resend email to suppliers; download tier gating; PDF report generation; online calculators (pultrusion selection / U-value); zh/en language switch; Clerk auth + enterprise claim flow.

STACK · Next.js 16 App Router · next-intl · Clerk · Resend · Vercel

MECHANICS

Four design choices that let the system self-run

LINK
Six libraries cross-link

Material pages show related formulas / recommended suppliers / downloads; paper pages show related papers; patent pages show related patents. Every page is an entry to the other five libraries.

CITATION
Citation-first AI

Every number and statement in an AI answer carries a [#N] citation that clicks to the source paper/patent/material page. When no source supports a claim, the AI says so explicitly — no hallucination.

BILINGUAL
Bilingual by design

next-intl path prefix /zh/ + /en/, default zh as-needed. One schema, two language renders, SEO hreflang bidirectional, llms.txt for LLM crawlers.

FLYWHEEL
Content flywheel

Daily cron: CrossRef/OpenAlex/USPTO ingestion + RSS news → AI Chinese commentary → multi-engine push (Baidu/Bing/Sogou/360) → Telegram channel → weekly newsletter. Self-running, no manual toil.

Now that you know the architecture, see what advantages it produces: