Intelligent Bidding Project Recommendation

Problem

Bidding opportunities are scattered across procurement notices, planned-construction records, attachments, and different project stages. Sales or pre-sales teams usually know their product, service, or solution, but they may not easily translate that into procurement language or identify which projects still have an actionable window. This skill addresses the gap between a business description and a shortlist of projects worth following up.

How It Works

It builds a business profile from minimal required and optional inputs, asking only one mandatory question when no product, service, or solution is provided. With a business name, it defaults to national scope, bidding records from the last 30 days, planned-construction records from the last 90 days, and no amount restriction, then calls the SBKJ bidding data API and pages until hasNext=false. Key steps include:
- Splitting the business into direct subject terms, procurement synonyms, scenario terms, demand-trigger terms, and exclusion terms, rather than relying on overly broad words such as “project” or “engineering.”
- Deduplicating, filtering, and verifying candidates using real list fields such as id and publishTime, then enriching only shortlisted items with details and source URLs when available.
- Scoring candidates on business match, opportunity stage, timeliness, region, amount, and evidence completeness, then producing A/B/C levels, evidence confidence, and reasons for excluded items.

Boundaries

API facts and analytical inferences must remain separate: returned project data is factual, while keyword expansion, scores, and follow-up advice are analyses. Missing region, amount, or customer type should not block a query, but the skill must not invent codes, contacts, budget truth, or win probability. If pagination, detail enrichment, or planned-construction source URLs are incomplete, the output should say so explicitly instead of filling gaps with guesses.