In Progress
Intelligent Bidding System
结合 AI 配置、RAG 与 Agent 的智能投标系统，用于提升文档处理和方案生成效率。
系统快照
架构、信号与证据会先于正文呈现。- Doc pipeline
- Structured
- Review loop
- Human-in-loop
Problem
投标场景涉及大量招标文件、历史方案、资质材料和格式要求。系统不仅要生成文本，还要理解文档结构、约束条件和业务事实。
Constraints
- 文档包含表格、图片、代码块和扫描件，不能简单按长度切分。
- 生成内容必须可追溯到来源。
- 关键结论需要人工复核，不能完全自动提交。
System Design
系统由文档采集、解析、分块、检索、重排、生成和审核流程组成。RAG 负责证据召回，Agent 负责多步骤任务编排，业务系统负责权限与状态。
Key Decisions
把投标系统做成“可控生成流程”比做成“万能写作助手”更可靠。每一步都需要中间结果、引用依据和人工确认入口。
Impact
项目目标是降低重复性文档整理成本，并提升方案生成的一致性和可复查性。
Reflection
后续重点会放在评测体系、模板管理、引用质量和多人协作流程上。
Problem
Bidding scenarios involve tender documents, historical proposals, qualification materials, and strict formatting requirements.
The system needs to understand document structure, constraints, and business facts rather than only generate text.
Constraints
Documents contain tables, images, code blocks, and scanned pages, so they cannot be chunked only by length.
Generated content must be traceable to sources.
Important conclusions need human review before submission.
System Design
The system contains document collection, parsing, chunking, retrieval, reranking, generation, and review workflows.
RAG recalls evidence, the Agent orchestrates multi-step tasks, and the business system manages permissions and state.
Key Decisions
A controlled generation workflow is more reliable than a universal writing assistant.
Every step needs intermediate results, citations, and an entry point for human confirmation.
Impact
The project aims to reduce repetitive document preparation and improve consistency and reviewability in proposal generation.
Reflection
Future work should focus on evaluation, template management, citation quality, and multi-person collaboration.