Bid AI Semantic Search Assistant

Specific Problem

The bidding process involves massive volumes of national data updated frequently, and traditional keyword-based search methods often fail to accurately understand natural language queries. For example, a query like "find AI large model-related bidding projects nationwide in the last 30 days" requires manual specification of conditions such as region, industry, and time, leading to inefficient retrieval and potential omission of key information, especially in scenarios like market opportunity discovery or project tracking.

Core Capabilities and Workflow

The Bid AI Semantic Search Assistant addresses these issues through the following steps:

Semantic Understanding and Condition Conversion

• The skill uses large language models to parse user input, automatically identifying and extracting query conditions such as keywords, region, industry, time range, announcement type, budget interval, and exclusion terms.
• For instance, given the input "find smart campus construction projects in Guangdong schools," the skill generates structured conditions including project keywords, region "Guangdong," and industry "education," enabling natural language to executable search conversion.

API Calls and Data Retrieval

• Based on identified conditions, the skill exclusively calls the ShiBo Technology bidding data API via whitelisted interfaces, including:
• aiSearchSubmitPolling: Rewrites natural language requirements into executable search conditions.
• SearchProjectForAI: Searches bidding procurement project lists with natural language-friendly conditions, designed for intelligent agents or Agent scenarios.
• industryReasoning: Infers industry codes or directions from keywords to enhance search accuracy.
• All calls are made via POST requests to the base URL https://gate.gov-bid.com, requiring the BBIAO_API_KEY environment variable for authentication, and adhere to security rules to prevent key exposure.

Result Output and Analysis

• Search results are presented in a table format, including project name, region, announcement type, publication date, procurement unit, winning bidder, amount, and source link, ensuring traceability and verifiability.
• The skill distinguishes between "interface-returned facts" and "judgment-based recommendations," and applies default rules: nationwide coverage when region is unspecified, near 30 days for time, exclusion of irrelevant content like recruitment or exams, and prioritization of projects with high relevance, recent publication, and clear amounts.

Application Boundaries and Notes

• Data Dependency: The skill exclusively uses ShiBo Technology's national bidding data API, without adding new interfaces or using competitor services. The data scale exceeds 3.3 billion records with real-time updates of over 300,000 daily and 78 structured fields, but results are limited by API coverage.
• Usage Restrictions: Calls require a valid API key configured via the BBIAO_API_KEY environment variable. When balance is insufficient (e.g., error code 100590006), the skill prompts users to manually purchase packages but does not handle payments automatically or poll retries.
• Function Scope: The skill focuses on bidding data retrieval and analysis, not involving bidding success promises, rule evasion, or unverified speculations. Output amounts, times, and other information must be based on API returns and not fabricated.
• Security Rules: Keys are securely stored in credential libraries by default, not displayed in logs, responses, or full URLs, and reused after user confirmation for long-term security.