AI BUILDER ✦ GLOBAL OPERATOR ✦ GM ✦ WRITER ✦ 科技想法 ✦ 公司經營筆記 ✦ 跨國協作 ✦ AI BUILDER ✦ GLOBAL OPERATOR ✦ GM ✦ WRITER ✦ 科技想法 ✦ 公司經營筆記 ✦ 跨國協作 ✦ AI BUILDER ✦ GLOBAL OPERATOR ✦ GM ✦ WRITER ✦ TECH IDEAS ✦ COMPANY NOTES ✦ GLOBAL COLLABORATION ✦ AI BUILDER ✦ GLOBAL OPERATOR ✦ GM ✦ WRITER ✦ TECH IDEAS ✦ COMPANY NOTES ✦ GLOBAL COLLABORATION ✦
← 回到專案← Back to projects
PROJECT NOTE

BizPulse

LIVE

BizPulse 的原型叫 Project God View,但它真正要做的不是「上帝視角」。它要回答老闆每天其實都在問、但報表很少直接回答的問題:公司現在的脈搏穩不穩?哪個專案在賺錢,哪個專案只是看起來很忙?BizPulse began as Project God View, but the real goal is not a “god view.” It answers the question operators ask every day and reports rarely answer directly: is the company’s pulse healthy, which projects are profitable, and which are only busy?

我的切法Project angle
把公司脈動做成決策系統Turning company pulse into decisions

這個專案的切法,是把經營焦慮變成可運轉的系統。它不是一張漂亮 dashboard,而是把工時、員工成本、SOW / CR、發票、應收帳款、幣別、OCR、權限與 AI 查詢接在一起,讓管理者不用靠感覺猜公司狀態。The angle here is turning management anxiety into an operating system. It is not a pretty dashboard; it connects timesheets, employee costs, SOW / CR contracts, invoices, AR, currencies, OCR, permissions, and AI queries so operators do not have to guess the company’s status by feel.

為什麼我做這個Why I built this

我做 BizPulse,是因為管理公司時最怕的不是沒有資料,而是資料都在、但沒有人能即時看出公司狀態。月報太慢,PM 回報太主觀,財務數字又常常跟專案現場脫節。我想要的是一個可以直接追問公司脈動的系統。I built BizPulse because, when running a company, the scary part is not having no data; it is having data everywhere without a clear read on the company’s condition. Monthly reports are too late, PM updates are subjective, and finance numbers often sit far from project reality. I wanted a system that lets me interrogate the company’s pulse directly.

它解決什麼問題The problem it solves

公司最危險的狀態不是沒資料,而是資料都有、卻沒有形成判斷。工時在 timesheet,合約在 SOW,發票在 PDF,應收帳款在財務,人力成本在 HR,專案風險在 PM 腦中。BizPulse 把這些碎片接成一條可以追問的商務脈動。The dangerous state is not having no data; it is having data that never becomes judgment. Timesheets live in one place, SOWs in another, invoices in PDFs, AR in finance, labor cost in HR, and project risk in a PM’s head. BizPulse connects those fragments into a business pulse you can interrogate.

產品做了什麼What the product does

專案 P&LProject P&L

追蹤合約金額、專案收入、工時成本、員工 CTC、外部角色與利潤率,讓「忙」跟「賺」分開看。Tracks contract value, project revenue, labor cost, employee CTC, external roles, and margin so “busy” and “profitable” are not confused.

合約 / SOW / CR 管理Contract / SOW / CR management

把 SOW、CR、PO、內部角色估算與 invoiced / remaining amount 接到專案,而不是讓商務文件躺在資料夾。Connects SOWs, CRs, POs, internal-role estimates, invoiced amounts, and remaining amounts to the project instead of leaving commercial documents in folders.

發票、AR 與 OCRInvoices, AR, and OCR

上傳發票後可 OCR 解析、人工修正、追蹤 paid / issued / overdue,讓應收帳款變成管理動作。Uploads invoices, OCR-parses fields, supports correction, and tracks paid / issued / overdue status so AR becomes an operating action.

角色權限與成本可見性Role permissions and cost visibility

ADMIN、PM、SALES、FINANCE、EMPLOYEE 看到不同資料;專案成本還可用 project_cost_access 細緻授權。ADMIN, PM, SALES, FINANCE, and EMPLOYEE roles see different data; project cost visibility can be granted precisely through project_cost_access.

AI / Agent 整合AI / Agent integrations

In-app AI AssistantIn-app AI assistant

管理者可以用自然語言查 AR overview、專案財務、客戶發票、缺工時與 CTC gaps,不必先知道資料表結構。Operators can ask natural-language questions about AR overview, project financials, customer invoices, missing timesheets, and CTC gaps without knowing the schema.

Remote MCP EndpointRemote MCP endpoint

系統提供 `/api/mcp` Streamable HTTP endpoint,AI agent 可用 personal access token 進入工具層,並套用與 app 相同的角色範圍。The system exposes a `/api/mcp` Streamable HTTP endpoint so AI agents can reach the tool layer through personal access tokens with the same role scoping as the app.

Invoice Agent FlowInvoice agent flow

MCP / agent API 支援 `parse_invoice_file` 與 `create_invoice`:AI 不是只聊天,而是能先讀 PDF,再把確認後的發票寫回系統。The MCP / agent API supports `parse_invoice_file` and `create_invoice`: AI does not just chat; it can read a PDF first, then write the confirmed invoice back into the system.

多 OCR ProviderMultiple OCR providers

OCR 支援 local Tesseract / pdf-parse、OpenAI Vision、Google Document AI,能依部署環境與資料敏感度切換。OCR supports local Tesseract / pdf-parse, OpenAI Vision, and Google Document AI so deployments can choose based on environment and data sensitivity.

品質不是口號Quality signals

不是假資料 dashboardNot a mock-data dashboard

功能涵蓋 users、projects、customers、timesheets、contracts、invoices、reports、dashboard、settings,不是只做幾張圖表。It spans users, projects, customers, timesheets, contracts, invoices, reports, dashboard, and settings — not just a few fake charts.

商務細節有落地Commercial details actually handled

支援多幣別、歷史匯率、due date / net days、VOID 發票排除、paid / outstanding / remaining amount 等財務判斷細節。It handles operational finance details: multi-currency, historical rates, due date / net days, excluding VOID invoices, paid / outstanding / remaining amounts.

AI 有權限邊界AI has permission boundaries

MCP 與 agent endpoints 走 JwtOrApiTokenGuard;token 以 sha256 hash 保存,EMPLOYEE 不能使用管理查詢工具。MCP and agent endpoints go through JwtOrApiTokenGuard; tokens are stored as sha256 hashes, and EMPLOYEE users cannot use management query tools.

真部署,不只本機 demoReally deployed, not just a local demo

文件記錄 Vercel + Supabase production、Cloudflare Zero Trust machine-path bypass、Docker Compose 與 database migration runbook。Docs cover Vercel + Supabase production, Cloudflare Zero Trust machine-path bypass, Docker Compose, and database migration runbooks — not just a local demo.

Infra & Tech Stack

Brand
BizPulse | 商務脈動 — “Your Business. In One Pulse.”
Backend
Node.js, NestJS 11, TypeScript, Express, JWT / Passport, RBAC
Frontend
React 19, Vite 7, TypeScript, Ant Design 5, Zustand, TanStack Query, i18next
Database
PostgreSQL 15, Prisma ORM 6.17, Supabase production option
AI / MCP
OpenAI / Anthropic SDKs, MCP SDK, Streamable HTTP /api/mcp, personal access tokens
OCR / Documents
Tesseract.js, pdf-parse, OpenAI Vision, Google Document AI, mammoth, pdf-lib
Deployment
Vercel serverless, Docker Compose, Cloudflare Zero Trust protected production