HackerNews中文版

AI 智能体在处理 B2B 数据时会产生幻觉，因为它们缺乏真实依据。我构建了一个这样的系统。 Nopp 的实体图谱从 NPPES 从业者数据、40 多个州执照委员会、ODO 公司注册处和监管备案中提取数据。没有抓取，没有推断。每个实体都经过确定性验证。我希望你进行压力测试的 API 接口：实体图谱 * POST /api/v1/identify — 姓名、NPI 或 LinkedIn URL → 经过验证的规范实体（10 IC） * GET /api/v1/people/:id — 获取已解析的个人记录（3 IC） * POST /api/v1/people/search — 按州、城市、证书筛选 1200 万+ 持证专业人士（10 IC） * GET /api/v1/organizations/:id — 获取组织记录（3 IC）情报 * POST /api/v1/company/enrich — 实时数据增强，优雅回退，永不返回空值（10 IC） * POST /api/v1/contact/enrich — 联系人级别的深度数据增强（3 IC）信号 * POST /api/v1/signals/hiring — 检测招聘激增（15 IC） * POST /api/v1/signals/funding — 融资轮次检测（15 IC） * POST /api/v1/signals/intent — 竞争对手的挫败感 + 意图信号（20 IC）智能体 * POST /api/v1/agent/launch/companies — 一次调用即可完成自主研究 + 外展（50 IC）也作为 MCP 服务器提供——直接连接 Claude、ChatGPT 或任何与 MCP 兼容的智能体。无需 API 调用，无需配置。我希望你做的事情： * 向 /identify 提供有歧义或冲突的输入——同一个人，两个 LinkedIn URL，不同的姓名拼写 * 尝试让 /company/enrich 返回空数据或错误数据 * 针对你非常了解的公司测试 /signals/intent — 它是否捕捉到真实的信号？ * 告诉我模式在哪里出错、文档不足或使用起来很烦人找到可复现的错误或真实的模式问题 → 我将给你免费的情报积分。无上限。 nopp.us/api-docs | nopp.us/developers 我是创始人。我会亲自回复每一个技术评论。

查看原文

AI agents hallucinate B2B data because they have no ground truth. I built one. Nopp's Entity Graph pulls from NPPES practitioner data, 40+ state licensing boards, ODO corporate registries, and regulatory filings. Nothing scraped. Nothing inferred. Every entity is deterministically verified. The API surface I want you to stress-test: Entity GraphPOST /api/v1/identify — name, NPI, or LinkedIn URL → verified canonical entity (10 IC) GET /api/v1/people/:id — fetch a resolved person record (3 IC) POST /api/v1/people/search — filter 12M+ licensed professionals by state, city, credential (10 IC) GET /api/v1/organizations/:id — fetch an org record (3 IC)IntelligencePOST /api/v1/company/enrich — live enrichment, falls back gracefully, never returns empty (10 IC) POST /api/v1/contact/enrich — contact-level deep enrichment (3 IC)SignalsPOST /api/v1/signals/hiring — detect hiring surges (15 IC) POST /api/v1/signals/funding — funding round detection (15 IC) POST /api/v1/signals/intent — competitor frustration + intent signals (20 IC)AgentsPOST /api/v1/agent/launch/companies — autonomous research + outreach in one call (50 IC)Also ships as an MCP Server — connect Claude, ChatGPT, or any MCP-compatible agent directly. No API calls, no plumbing. What I want from you:Feed /identify ambiguous or conflicting inputs — same person, two LinkedIn URLs, different name spellings Try to get /company/enrich to return empty or wrong data Test /signals/intent against a company you know well — does it catch real signals? Tell me where the schema is wrong, underdocumented, or just annoying to work withFind a reproducible bug or a real schema issue → I'll give you free Intelligence Credits. No cap. nopp.us/api-docs | nopp.us/developers I'm the founder. I'll reply to every technical comment personally.

我为 AI 智能体构建了一个身份图谱——包含 3.3 亿+ 经过验证的记录。 停止 API

我为 AI 智能体构建了一个身份图谱——包含 3.3 亿+ 经过验证的记录。停止 API