English Introduction

AI Project Advisor is a WeChat-based AI assistant for construction / engineering / building projects. It helps project owners and managers run projects by just chatting in WeChat groups — no software to learn, no forms to fill.

Key Features

• Memory Graph (Chat = Archive): Every word in your project group is archived automatically — promises, confirmations, payments, changes — with who/when/original message, ready as evidence for payment collection and dispute resolution
• Multi-Group Support: One project per group (WeChat Work group), auto-binding group ↔ project, no data interference between projects; boss chat = global view
• Rule + LLM Hybrid Engine: Daily recording costs zero tokens (rule-based intent recognition); complex understanding / deep diagnosis / report generation use LLM on demand (DeepSeek — cheap)
• Risk Radar: Litigation / key personnel changes / supplier failures — proactive alerts + action checklists (auto web-scan is a paid-tier feature)
• 8 Advisor Modules: Progress / Payment / Change / Documentation / Risk / Inspection / Review / Training
• Proactive Service: Daily morning brief, deadline countdown, anomaly detection (overdue payments, missing materials, schedule slippage), one-click reports, voice/photo input

Quick Start

# Python 3 + optional DEEPSEEK_API_KEY (in ~/.hermes/.env or env var)
python3 advisor_engine.py   # multi-group demo
python3 advisor_llm.py      # LLM enhancement test (diagnosis/report)

New in latest version

• WeCom Smart Bot (企微智能机器人 API 模式): FREE channel — receive files/text/images from users, encrypted callback reply (msgtype=stream), file decryption (AESKey)
• File Auto-Project (文件智能建项目): Upload .xlsx / .xls / Word(.docx) / text — AI extracts project name, contract amount (auto from 合计 row), duration → auto-create project + milestone plan
• Contract Risk Assessment (合同风险评估): contract text stored on upload → risk advisor analyzes contract clauses / client / project in 3 dimensions with P0/P1 action plan
• Photo-to-Ledger (拍照即记账): receipts / visa / acceptance photos OCR → auto-record business facts
• Speech-to-Data (说话录入): "收到进度款20万" → auto-update received amount; changes / promises / issues / progress all auto-recorded
• 8 Advisor Modules (八大顾问): Progress (critical path + catch-up) / Payment (aging + collection) / Change (visa completeness) / Docs / Risk (contract-aware) / Inspection / Review / Training — each with professional framework
• Resource Management (资源方管理): supplier/team profile, arrival/shortage tracking, efficiency & load alerts
• Memory Recall (记忆图谱): "回忆一下XX说过什么" — query by person/topic from archived originals
• Business Decisions (经营决策): profit forecast / cashflow gap / receivable aging / bid evaluation
• Proactive Morning Brief (主动预警): daily 7:30 cron — milestone due / overdue payment / open risks
• Guided Onboarding (引导性建档): AI walks the client through 6-step base-doc setup one step at a time
• Visa Ledger (签证台账, 2026-08-23): "签证：增项吊顶（2.5万）" register → "签证清单" query (pending/signed/paid) → "签证：已签 XX" update; change advisor reads the ledger
• Material Ledger (材料台账, 2026-08-23): "材料：海尔冰箱 100台 1750元" register (name/qty/price parse) → "材料清单" query → "材料到货：XX" arrival update
• Milestones by Project Type (节点按工程类型, 2026-08-23): auto-detect weak-current/decoration/installation/civil works from project name → 4 stage templates with real construction flows
• Industry Knowledge Base (行业知识库 L1, 2026-08-23): 8 categories × 12 entries with legal citations (民法典 807, 司法解释一, 建质〔2017〕138号...) — direct Q&A ("质保金一般留多少") or auto-injected into advisor analysis
• Fix: "付款节点" no longer misread as milestone query (付款节点误判修复, 2026-08-23)
• Milestone Plan (节点计划): 5-stage schedule with deliverables + progress bar + overdue tracking

Architecture

Chat (WeChat Work group)
  → Rule engine (intent recognition — 0 token)
  → Memory graph (SQLite: projects/events/group_bindings...)
  → LLM layer (DeepSeek: complex parsing/diagnosis/reports — on demand)
  → Proactive push (morning brief / alerts / reminders)

Free vs Paid

• Free: 3 project groups + recording/reminders + limited daily Q&A
• Paid: more groups + deep diagnosis / reports / risk radar auto-scan

MIT License. Star & share if useful!

---

🏗️ 工程项目 AI 顾问（AI Project Advisor）

微信对话式工程项目管理助手——老板不用管项目，AI 替他管：发现问题、给出方案、自动干活。

核心能力

• 记忆图谱（对话即档案）：群里说的每句话自动存档——承诺/确认/付款/变更——有据可查（清款依据）
• 多群接入：一个项目一个群——群↔项目绑定——多项目互不干扰
• 规则 + 大模型混合：日常记录零成本（规则引擎），深度诊断/周报按需调大模型（DeepSeek）
• 风险雷达：官司负面/人员变动/资源方暴雷——预警+核查清单（自动扫描=收费版）
• 8 大顾问模块：进度/回款/变更/资料/风险/巡检/复盘/新人

快速开始

# 依赖：Python 3 + 可选 DEEPSEEK_API_KEY（~/.hermes/.env 或环境变量）
python3 advisor_engine.py          # 跑多群演示
python3 advisor_llm.py             # 测大模型增强（诊断/周报）

使用示例

老板（群里）：新建项目：XX大厦办公装修，合同80万，已收30%，负责人王经理
AI：✅ 项目已创建，群已绑定——群里说的自动记录

老板（群里）：刚收了进度款20万
AI：✅ 已记录收款 20 万，累计已收 44 万（合同 80 万）

老板（群里）：甲方张经理离职了，换了个李工
AI：⚠️ 人员变动核查清单：签证补签/承诺书面化/增项补签...

老板（群里）：生成周报
AI：📋 完整周报（进展/问题/回款/计划/风险）

架构

对话（企业微信/群）
  → 规则引擎（意图识别——0 token）
  → 记忆图谱（SQLite：projects/events/group_bindings...）
  → 大模型层（DeepSeek：复杂句/诊断/周报——按需）
  → 主动推送（早报/预警/提醒）

免费版 vs 收费版

• 免费版：3 个项目群 + 记录/提醒 + 每天限量问答
• 收费版：更多群数 + 深度诊断/周报/风险雷达自动扫描

文件

• advisor_engine.py — 核心引擎（意图识别/图谱/多群/预警）
• advisor_llm.py — 大模型增强层（DeepSeek）
• 记忆图谱技术设计.md — 数据库设计
• 产品方案书.md — 完整产品设计

MIT License