Bid-Studio: Intelligent End-to-End Bid Document Assistant

The Specific Challenges in Bidding

The bidding process is crucial for winning contracts, but it faces two core pain points: first, the creation of compliance documents is time-consuming and error-prone, with strict formatting for bid letters, technical proposals, and price sheets where any detail oversight can lead to disqualification. Second, the risk of bid collusion or similarity is covert; unintended similarities or organized likenesses across multiple bid documents are hard to detect manually, making manual comparison inefficient and risky. These challenges demand a tool that both generates standardized documents and performs in-depth semantic risk analysis.

The Skill's Dual-Line Workflow

This skill addresses these issues through two independent yet connected workflows: Line A (Creation) and Line B (Screening).

Line A: Bid Document Generation & Compliance Self-Check

The core of this module is translating industry standards into executable rules for the AI. It embeds a "Rule Set" that mandates providing critical information like project name and company full name, and strictly prohibits fabricating qualifications. The generation process involves:
1. Template-Based Filling: Users select a template (e.g., "Engineering," "Goods Procurement," or "Service Project"). The AI generates content following a fixed directory structure (e.g., bid letter, technical proposal, price proposal) based on the provided company data and project info.
2. Built-in Guardrails: During generation, the AI automatically performs a "Disqualification Risk Self-Check," verifying 32 red-line items such as consistency of price figures (numeric and Chinese characters), notation of negative deviations, and completeness of mandatory clauses.
3. Output & Iteration: Output is in Markdown format. A set of modification commands is provided for refinement (e.g., "continue," "polish," "rewrite"), supporting fine-grained adjustments to sentences, paragraphs, or sections.

Line B: Collusion & Similarity Risk Screening

This module combines deterministic scripts for document parsing with Large Language Models for semantic analysis.
1. Structured Extraction: The extract_documents.py script parses uploaded .docx/.pdf/.txt files into a unified JSON structure containing paragraphs, tables, and metadata.
2. Intelligent Detection: The LLM analyzes the extracted content based on a built-in "Sensitive Field List" and "Similarity Grading Criteria." Detection is prioritized: Critical Information Collision > Document Attribute Comparison > Text Semantic Similarity > Table Structure Similarity.
3. Report Generation: The build_report.py script renders the analysis results (findings.json) into a structured screening report, which includes a risk summary, detailed pinpointing, and a comprehensive conclusion. Personal information is automatically anonymized in the report.

Applicable Boundaries & Notes

This skill is an assistive tool with clear use cases and boundaries:
- Line A: Suited for常规-sized bids (recommended for under 500 pages per run). It does not support generating legally binding stamped documents, nor does it query real-time bidding information or "guarantee a win." All output must be manually reviewed before use.
- Line B: Positioned as a preliminary screen and decision aid. Its report does not replace professional plagiarism systems, the official judgment of a bid evaluation committee, or legal determinations. It excels at detecting explicit similarities and structural likenesses but has limited capability against deeply rewritten or reformatted content.
- Workflow Closure: It is recommended to export the document from Line A and use it as input for Line B, cross-referencing with historical bids or the tender documents to proactively mitigate risks.
- Privacy & Security: Processing does not persist data. Reports automatically anonymize private information. Temporary files should be cleaned up after the session.