guangshu100/BidMaster-Pro

BidMaster Pro – AI‑powered end‑to‑end tender (招投标) platform

What it is

BidMaster Pro is a desktop‑plus‑server application that automates the whole tender‑to‑bid workflow with large language models (LLMs). It chains together four stages – Interpret (tender parsing), Generate (bid drafting), Check (compliance review), and Format (document layout) – using an Agent + Skill engine built on LangGraph/LangChain. The system can switch between multiple LLM providers (DeepSeek, OpenAI, Alibaba Tongyi, SiliconFlow, local Ollama models) and augments generation with a RAG knowledge base (vector‑store of company documents and past bids) and MinerU OCR for scanned PDFs.

Core capabilities

Stage	Main functions
Interpret	Parse PDF/DOCX/TXT tenders, extract scoring criteria, qualifications, technical specs, risk flags, and produce a scoring matrix.
Generate	Auto‑create an outline, write each section with LLM assistance, retrieve relevant past cases from the RAG store, and generate supporting charts/diagrams.
Check	Run 21 predefined compliance checks (e.g., bond, signatures, validity dates, consistency, plagiarism, price reasonableness) and flag any violations.
Format	Apply a chosen bid template, auto‑style the document, and export as DOCX, DOC, or PDF. Supports diff against the template and collaborative revision.
Auxiliary	• AI‑driven Q&A chat that cites source fragments. • Tender‑monitoring service that scrapes public tender sites, filters by keywords, and scores opportunity value. • RBAC for multi‑user teams.

Architecture at a glance

Desktop client (React + Electron + TypeScript)
   ⇆ REST API ⇆ FastAPI server (Python 3.12+)
        ├─ Router layer (projects, interpret, generate, …)
        ├─ Core engines
        │   • Agent Engine (LangGraph) – orchestrates Skills
        │   • Skill Engine – plug‑in business logic
        │   • RAG Engine (ChromaDB) – vector search & semantic re‑ranking
        │   • LLM Gateway (LiteLLM) – unified access to many LLM providers
        │   • Document Engine (pdfplumber, python‑docx, etc.)
        └─ Infrastructure
            • PostgreSQL (asyncpg) – relational data
            • Redis + Celery – task queue & caching
            • MinIO – object storage for uploaded files

Main tech stack

• Backend: FastAPI, SQLAlchemy, asyncpg, PostgreSQL 16, Redis 7, Celery, MinIO
• AI/LLM: LiteLLM (multi‑provider), LangGraph, LangChain‑Core, ChromaDB, Sentence‑Transformers, ONNX Runtime for local models
• Frontend: React 19, Electron 41, TypeScript 5, Vite, TailwindCSS, Zustand, TanStack Query, Radix UI
• Document handling: python‑docx, pdfplumber, PyMuPDF, Mammoth, WeasyPrint, BeautifulSoup4

Getting started (quick‑start)

1. Clone & configure – copy .env.example to .env and fill in LLM API keys, DB URLs, etc.
2. Docker deployment (recommended)

   docker-compose up -d   # brings up Postgres, Redis, MinIO, FastAPI, Celery
   

   API docs are then at http://localhost:8000/docs.
3. Local development (if you prefer to run the code yourself)
   • Create a Python venv, install the package (pip install -e .).
   • Run migrations (alembic upgrade head).
   • Start FastAPI (uvicorn services.main:app --reload).
   • Launch Celery worker (celery -A services.celery_app worker).
   • In packages/desktop install Node deps and start the Electron app (npm run electron:dev).

Extensibility

• Skills are Python classes that encapsulate a single capability (e.g., outline generation, OCR extraction). New Skills can be added and wired into the LangGraph pipeline via a JSON‑like DSL.
• Agent pipelines define the order of Skill execution; a Gate Keeper ensures earlier steps finish before later ones start.
• LLM provider switching is handled by LiteLLM, so you can add any provider that follows the OpenAI‑compatible API.

Who might use it

• Companies that regularly respond to public tenders and need to speed up bid preparation while ensuring regulatory compliance.
• Consulting firms building internal bid‑automation tools.
• Developers looking for a reference implementation of a multi‑LLM, RAG‑backed workflow with a desktop UI.

---

License: MIT – see LICENSE. Contact: Scan the QR code in the README to follow the WeChat public account and message bidmaster for author details and support.