reddit · SideProject · 09-29 19:50 50 分 ← 返回列表

Why naive RAG failed us in custom manufacturing (and how we built an auditable memory layer with Hindsight & Groq)

Hey everyone, Our team, HackOrbit, recently wrapped up a hackathon project called OrderAnchor, an open-source system designed to tackle a problem that can constantly cause costly mistakes in B2B manufacturing and custom apparel: communication drift across fragmented channels. The Problem In custom print and apparel shops, orders rarely arrive as clean database records. They come through WhatsApp conversations, forwarded emails, phone calls, and multiple rounds of revisions. For exampl

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：定制印刷/服装加工店的订单散落在 WhatsApp、转发邮件、电话和多轮修改中，语义相似的旧消息常被误当成最终确认，导致错料、错产和真金白银的损失。

目标用户：中小型定制印刷/服装加工厂（B2B 制造）的老板、订单协调员与生产主管

证据：The result can be wasted fabric, incorrect production, and significant financial loss. / In physical production, a confident but unverified answer can be much worse than no answer at all. Plausible does not mean verified.

变现：垂直行业 SaaS 订阅：按席位或按订单量阶梯收费（如 $99–299/月/门店），可加人工审核工作流、审计日志、与现有 ERP/印刷管理系统的付费集成。

MVP：2 周 MVP：接入 WhatsApp/邮件导入，用 LLM 把消息抽成「提案/修改/附加条件/最终确认」状态机，任何写入生产订单前必须人工点确认，并保留可审计的变更时间线；先只做 1 个品类（丝网印刷 T 恤）和 1 个渠道。

竞品：Printavo、Teesom、通用向量 RAG 客服机器人（Intercom/Zendesk AI）、通用 ERP/订单管理系统

摘要：HackOrbit 团队在黑客松中做了开源项目 OrderAnchor，针对定制印刷与服装加工行业「跨渠道沟通漂移」问题：订单来自 WhatsApp、转发邮件和多轮修改，朴素向量 RAG 会把语义最相关的旧消息误判为最终确认，从而生产出错、浪费面料并造成重大财务损失。方案是用 FastAPI + SQLite + Groq(Llama 3) + Vectorize Hindsight 搭建带人工复核的审计型记忆层。帖子本身是技术分享，没有真实客户或收入数据，但痛点场景和损失描述具体，适合作为垂直订单确认工具的机会起点。