Bid Natural Language Search Assistant

Problem

Bidding searches often get stuck on field mapping: a user may know “school air-conditioner procurement wins in Guangdong,” but cannot express API parameters; engineering teams still need to normalize time, region, notice type, exclusions, and missing fields. This skill turns conversational questions into executable search conditions and uses the existing Shibo Technology bidding data API to retrieve announcements, awards, contracts, pre-construction projects, files, and source links.

How It Works

• Intent detection: distinguishes general search, project number lookup, details, contracts, pre-construction, attachments, or industry reasoning.
• Condition extraction: identifies keywords, region, industry, buyer, amount, time, notice type, and exclusions.
• Conservative defaults: if region is missing, use national scope; if time is missing, default to the last 90 days; include procurement intent, tender notices, award results, and contract notices.
• AI rewriting: for full-sentence or multi-condition queries, prefer aiSearchSubmitPolling; for industry phrases, use industryReasoning to infer candidates.
• API retrieval: query with SearchProjectForAI or searchProjectApi, then call detail, attachment, and source-link endpoints as needed.

Boundaries

This skill does not add new data endpoints and only uses the listed existing endpoints. Output depends on API responses. Missing fields should be marked “not disclosed” or “unknown”; do not invent amounts, contacts, attachments, or links.