Bidding Compliance Review Engine

Problem

Risks in tender documents are often spread across eligibility conditions, technical parameters, scoring methods, contract clauses, and bidding procedures. Manual review can miss items, and it is hard to trace each finding to the original text, the corresponding review item, and the regulatory basis.

How it works

The skill codifies review rules into 8 YAML rule files with 95 rules in total, then executes them with a Python rules engine. The goal is to make the review process auditable, reproducible, and versionable.
- Accepts PDF, DOCX, and TXT input, extracts text, and uses heuristic section recognition to produce a structured tender.json.
- Scans eight dimensions in sequence: fair competition, disqualification clauses, qualification conditions, technical parameters, bidding and evaluation procedures, contract clause risks, contract and procedure compliance, and antitrust or special-review items.
- Scans append results instead of overwriting earlier findings; each issue records clause location, evidence, remediation advice, regulatory basis, and check_item identifier.
- Produces a review conclusion summary, a high-risk issue list, and a downloadable report for follow-up remediation and archiving.

Scope and caveats

• Rule coverage is limited to the listed review items; details outside the rule set still require manual review.
• Scanned PDF input depends on OCR quality, and poor section recognition can reduce the granularity of issue localization.
• The workflow must run phase by phase and stop on failure, using audit logs for traceability. It is not a substitute for real-time legal research or issuing a formal legal opinion.