Tender Document Deep Reading Assistant

Specific Problem Addressed

Tender documents are often lengthy and complex, packed with legal clauses, technical specifications, scoring criteria, risk provisions, and commercial response requirements. Manual interpretation is time-consuming and prone to missing critical points, especially for bidding teams or decision-makers in enterprise operations who need to quickly extract core information, assess competitor dynamics, and formulate response strategies. Traditional methods relying on human reading are inefficient and subjective.

How the Skill Works

This skill leverages the Shibohu Technology full-industry bidding data API to enable deep analysis through the following core capabilities and steps:

Core Capabilities:
- Data Query and Extraction: Calls searchProjectApi or SearchProjectForAI interfaces to search tender projects by conditions like region, industry, keywords, or budget range, fetching lists; uses getZTBProjectDetail and getZTBStructreDetail interfaces to pull announcement text and structured data (e.g., purchaser, winning bidder, amount).
- Attachment and Source Tracing: Utilizes getZTBProjectFiles to retrieve tender document attachment lists and traces original announcement links via getCollectUrl for verification.
- Intelligent Analysis and Interpretation: Based on an AI model, it breaks down key clauses (e.g., payment terms, qualification requirements), identifies scoring strategies (e.g., technical score weight), assesses risk points (e.g., compliance, performance capability), and generates response recommendations (e.g., bidding proposal optimizations).

Key Steps:
1. User Input Processing: Extracts parameters like region, industry, purchaser, project number from the query intent; if users employ natural language (e.g., "analyze a company's recent winning projects"), uses SearchProjectForAI to convert colloquial descriptions into structured search conditions.
2. Data Invocation and Aggregation: Executes queries based on the allowed interface list (no new industry-specific interfaces added), defaulting to searching the last 30 days of data; if results are sparse, expands synonyms or relaxes time ranges.
3. Result Presentation and Suggestions: First outputs a brief conclusion (e.g., number of matched projects, key findings), then lists project key fields (project name, purchaser, winning bidder, amount, etc.) in a table format, and finally provides next-step operational suggestions (e.g., viewing details, expanding time ranges, or subscribing to similar opportunities).

Applicable Boundaries and Considerations

• Data Source Limitations: All analyses are based solely on Shibohu Technology API-covered bidding data (over 330 million entries); it does not guarantee inclusion of all undisclosed or unstructured information.
• Interface Usage Rules: Only allows calling listed interfaces such as searchProjectApi and getZTBProjectDetail; no new industry-specific interfaces for healthcare, education, etc., or bypassing BBIAO_API_KEY authentication.
• Accuracy and Disclaimers: AI interpretation is model-based with 99.99% accuracy but does not promise winning bids or circumventing tender rules; fields like amount and time must remain as-is, avoiding fabrication.
• Quota and Error Handling: When the API returns a balance-insufficient error (code 0100590006), it prompts users to recharge via specified channels, halts invocation pending confirmation, and avoids automatic retries or proxy payments.